Compare commits

..
Author SHA1 Message Date
BZLZHH 7fd7f71d6c [Feature] (EGL): Wayland window surfaces, presented through wl_shm
A Wayland client's window is its own wl_surface, and what the compositor shows is the wl_buffer
attached to it - but the pixels of a MobileGL frame are drawn where the backend is, on a split
build in another process on another OS.  So far a wl_egl_window was taken for an X11 window and
its client-side address sent to the Android server (EGL_BAD_NATIVE_WINDOW at best), and under
glvnd a Wayland display never reached this vendor at all.

A window on a Wayland display - asked for as EGL_PLATFORM_WAYLAND, or a wl_display recognised by
its first word being wl_display_interface - is now a window surface in the EGL state at the
wl_egl_window's size with a pbuffer of that size behind it in the backend, whatever
MOBILEGL_IPC_SURFACE says.  At every swap the frame is read back out of the default framebuffer
through this library's own glReadPixels (the application's read framebuffer, pack buffer and pack
state put back as they were), its rows flipped and swizzled into a wl_shm buffer, and attached,
damaged and committed on the application's wl_surface.  Three buffers rotate on their release
events, dispatched on an event queue of the surface's own.  libwayland-client is loaded at run
time, so neither the build nor the library depends on it.  EGL_KHR_platform_wayland and
EGL_EXT_platform_wayland are announced on Linux.

Measured: glmark2-wayland runs its build scene at 13 FPS (400x300) and the window shows the frame
under kwin.  A window resize is not followed yet; the surface keeps its creation size.
2026-10-01 10:59:07 -04:00
BZLZHH 353daa265e [Fix] (Pipe): the records two families keep of their own are forgotten with the session too
MGPipeForgetAllPublications cleared the publication latch, but two death helpers do not ask the
latch: vertex elements ask the vertex-input emitter's RecordIsPublished and buffers ask the
resource tracker's WasPublished.  A VAO and its buffers destroyed with the GL context after the
split roles were down still found themselves published there, and glmark2's eglTerminate stopped
on the same assertion eglinfo's did (ResolveVertexElements, then ResolveResourceIn for a buffer).

Both records are now dropped beside the latch: the emitter's latches whole, the tracker's
Published flags.
2026-10-01 10:59:06 -04:00
BZLZHH 69728a4f9b [Fix] (Remote): the create window starts empty with each session
The window of creates whose answers are read later is process-global, and every entry names a
sequence number of the session that wrote it.  A process can run several sessions one after
another - eglinfo terminates one platform display and initializes the next - and an entry left by
the previous session made the new session's first create wait, in DrainCreateWindow, for a
sequence number the new server would not reach: the second session hung on its first texture
create.

InstallClientWireTables, which runs at the start of every session, now empties the window and the
create suspects; their answers and their objects died with the session that wrote them.
2026-10-01 10:31:20 -04:00
BZLZHH c6f828532a [Fix] (Pipe): nothing is published once the split session is gone
MobileGL::Destroy shuts the split roles down first and resets the GL context after, by design.  The
frontend objects that die in that reset still found their handles published, so their deletes were
routed to whichever applier was installed by then - the monolith one, which never had those
resources and asserted ("ResourceDestroy named {slot, gen}") from eglTerminate.  The abort also
threw away the program's buffered output, so eglinfo printed nothing but the assertion.

Every twin the server held died with the session, so ShutdownSplitRoles now clears the publication
latch (MGPipeForgetAllPublications): the deletes that follow are local bookkeeping only.
2026-10-01 10:31:20 -04:00
BZLZHH 5bb28140cb [Feature] (Remote): the TCP supervisor serves sessions side by side
A desktop is a compositor AND the GL programs run inside it, each a client with a session of its
own.  The TCP supervisor answered every authenticated Hello after the first with Refuse{Busy}, so
with kwin connected any other program - eglinfo, glxinfo, glmark2 from konsole - waited out its
Welcome and died on Fatal{CapsBeforeFirstSnapshot}.

The fork shape now keeps a list of live children and forks for a Hello while fewer than eight are
live.  What made it a single slot was the data connection: a DataBind names its session only by
the data nonce, which the child minted, so the supervisor could hand it to nothing but "the" child.
The nonce is now minted by the supervisor just before fork(); the child adopts it
(PresetDataNonceForThisProcess) instead of minting its own, and a DataBind goes to the child whose
nonce it presents.  Nothing on the wire changes.  The thread shape (the in-process display server)
still serves one session, Busy for a second as before.
2026-10-01 10:31:20 -04:00
BZLZHH bf7fc6bf6b [Fix] (DirectGLES): an imported external texture is attached as the host frame canvas
The arms that bind the display host's frame in place of an imported texture were gated on the
attach target being 0x8D65, but an external texture is attached through the GL_TEXTURE_2D upload
target, so they never ran.  The texture's twin on the server has no storage - the dma-buf behind it
never crosses - and kwin's framebuffers over it were incomplete (status 0x8cd7, every draw
GL_INVALID_FRAMEBUFFER_OPERATION), so nothing it drew landed anywhere.

The record arm (SyncAttachmentSurface) now recognises a texture point whose target is
TextureExternalOES and attaches the host frame canvas before any twin adoption or storage sync,
which have nothing to offer a texture with no storage; it takes the host frame first if none is held
yet.  The legacy object arm keys on the texture object's own target the same way.  Measured: all
four of kwin's daemon-buffer framebuffers complete, and the desktop it draws reaches the glass.
2026-10-01 09:47:32 -04:00
BZLZHH b72d4c5dff [Feature] (DirectGLES): a stable canvas in front of the host frames
Every present hands the frame back to the display host and takes a different one, and the frame's
renderbuffer is deleted with it.  A compositor attaches its framebuffers once (kwin: one per daemon
buffer, for the life of the process), so attachments bound to a frame's own renderbuffer named a
deleted object from the second frame on.

The backend now keeps a canvas - a renderbuffer and framebuffer of its own, made at the first
frame's size and kept for the session.  The host-frame globals (the default-framebuffer redirect,
the storage external attachments are bound to) name the canvas instead of the frame, and keep
naming it between a frame being handed back and the next one arriving.  PresentHostFrame copies the
canvas into the frame before its fence; the copy puts back the compositor's framebuffer bindings,
scissor test and rasterizer discard exactly as they were.  Without a canvas the frame itself stays
the draw target, as before.
2026-10-01 09:47:24 -04:00
BZLZHH c88c6ff196 [Fix] (Remote): the present record presents a host-framed surface
A split client's swap reaches the server as a present record, and ServerVerbSink::OnPresent only
ran the backend's Present - an eglSwapBuffers of the pbuffer the context was made with.
BackendObject::SwapEGLBuffers is the only route to PresentHostFrame and a split client never takes
it, so over TCP the display host was never told a frame was drawn: the app offered thousands of
frames and presented none.

BackendObject::PresentHostFrameIfCurrent presents the host frame when the current surface is
host-framed and is a no-op otherwise; OnPresent calls it after Present and before the credit goes
back, so a client released by the credit is released by a frame that is on its way to the glass.
2026-10-01 09:47:18 -04:00
BZLZHH 16488d5801 [Feature] (GL): glFlush presents the display host's frame for a compositor that never swaps
kwin draws the desktop into the display host's frame and then calls glFlush - and glFlush alone;
it never calls eglSwapBuffers.  The stream reached the server and the server drew, but the display
host was never told a frame was done, so nothing reached the glass.  When the surfaceless-bind hook
has taken the host frame, glFlush now swaps that surface after flushing the stream: that swap is the
host's "frame is done".

A swap of a surface the EGL state does not know used to return EGL_BAD_SURFACE without a word - the
whole reason the host frame's presents were invisible - so that refusal is now reported once.
2026-10-01 09:47:12 -04:00
BZLZHH c16a51da62 [Feature] (EGL): a surfaceless compositor takes the display host's frame surface
A compositor on the anland route never creates a window surface: it imports the display host's
dma-bufs, draws into them through framebuffers and binds its contexts with EGL_NO_SURFACE.  The
host frame bridge is only opened by a host-framed surface being created and activated, so with no
surface the bridge stayed shut and the display host offered its frames into a name nobody served.

So the first surfaceless bind of a real context, in a session that asked for host-framed surfaces,
makes the surface eglCreateWindowSurface would have made - registered in the EGL state AND created
through the backend's window-surface entry, which the remote backend routes to the host frame - and
every later surfaceless bind binds it, because the compositor rebinds with EGL_NO_SURFACE all the
time and a bind without it leaves nothing current to present.  Registering it in the EGL state is
not optional: a backend-only token was refused by the state (EGL_BAD_SURFACE) on every bind and
every swap.

Only the compositor takes the host frame.  Desktops are full of processes that bind with no surface
(konsole, plasmashell, kactivitymanagerd), and a second process taking it made the activation fail;
the compositor's launcher marks it with MOBILEGL_HOST_FRAME_CLIENT=1.
2026-10-01 09:47:06 -04:00
BZLZHH 08e45dc9c6 [Fix] (GLState): an external texture is a 2D texture that remembers it is external
GL_TEXTURE_EXTERNAL_OES had a target of its own but nothing behind it: the texture factory had no
row for it (the compositor died on "Unimplemented texture type" the moment it bound one), the
attach path had no upload target for 0x8D65 (validation failed, the attachment was never recorded
and the compositor read GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT), and nothing downstream
could tell an imported texture from an ordinary one.

An external texture is now a TextureObject2D created with the TextureExternalOES target, attached
through the 2D upload target - which the three framebuffer-texture validations accept for an
external object - and the target crosses the pipe (TexExternalOES maps back on the server) so the
render server can recognise it.  To the driver it stays a 2D texture: MapToBackendTextureTarget
folds it to Texture2D like the other targets ES does not have.
2026-10-01 09:46:57 -04:00
BZLZHH b2ce268c86 [Fix] (GLState): a global buffer target asked through the indexed lookup is a warning, not an abort
Only four buffer targets have per-index binding points (Uniform, TransformFeedback, AtomicCounter,
ShaderStorage).  A caller that asks GetBindingPoint for any other target is reading the wrong
accessor, not passing a bad enum - and kwin reached exactly that right after its vertex-attrib
setup, where the assertion killed the whole compositor with Trace/breakpoint trap.

The answer is now a one-time warning and index 0 of the indexed array: the caller gets a valid
slot, the global binding is untouched, and the mismatch stays visible in the log.
2026-10-01 09:46:46 -04:00
BZLZHH df58f21d83 [Feature] (EGL): the vendor answers what a compositor resolves, and the imported texture is complete
A compositor's bring-up is a fixed sequence of questions that all have to be answered by the same
library: which devices exist, which display a device serves, what its DRM node is, which platform
the display comes from, which configs exist, a context, a surfaceless make-current, the GL entry
points for that context, and finally the dma-buf import it draws with.  Every missing answer in that
list cost a run to find, and each is measured rather than assumed - an unanswered device query was a
call through a null pointer, an unanswered GL name left glvnd's dispatch table empty so glGetString
answered NULL through libGL while the same context answered through eglGetProcAddress, an unanswered
node string produced "couldn't find dev node for drm device", and a rejected
GL_TEXTURE_EXTERNAL_OES binding is why the framebuffer over an imported dma-buf was reported
incomplete.

The EGL side is whole now: the device-query family is published as real entry points with the DRM
node named, the dispatch hook answers GL names as well as EGL ones, and the swap-and-sync family the
compositor checks between its context and its first surface - EGL_KHR_fence_sync, EGL_KHR_wait_sync,
the damage variants, EGL_EXT_buffer_age - is implemented and announced.  A surfaceless context is
recorded as what it is rather than as a surface that does not exist.  These entry points log what
they are asked at DEBUG level, because the next bring-up should not need a disassembler.

GL_TEXTURE_EXTERNAL_OES is a target of its own (its TextureTarget, its pipe row, the GLenum mapping,
and the OES_EGL_image extension named), which is what lets the compositor's bind succeed at all; the
texture an imported image backs is marked backed, so a framebuffer built over it is complete.  The
client half of the import is new beside it: each plane's descriptor is described and shared with the
render server over the session transport, which is the change the server side has been waiting for.
2026-10-01 18:15:55 +08:00
gold 29a89cc10a [Fix] (gbm): the two defects only the Android build could find
Neither showed up in the Linux build that verified this library, and both are the kind that would
have been found by the first person to build the phone half - so they are fixed here rather than
left for that person.

bionic only declares memfd_create from API 30 and this library builds against 26. The kernel call is
far older than the declaration, so it is asked for by number: the same call with the name the
headers of this API level do not carry yet.

The library carries the tree's logger, which reaches Android's own log through liblog. The main
target gets that library in its own Android block; a second target built from the same sources needs
it for the same reason, and without it the link is what fails rather than anything in the code.
2026-10-01 08:23:04 +08:00
gold 8e3db2cb87 [Feature] (gbm): the server offers each taken frame to the container
The channel existed and had no callers.  Now the frame the server takes from the host is offered to
the container with the geometry the host decided, the apply loop consumes the releases the
compositor sends back, and the teardown that drops the host frame closes the channel.

Every call is on the optional path: with no container the render path is what it was.
2026-10-01 03:42:18 +08:00
gold 1ba7b1a96f [Feature] (gbm): the server sends each frame's dma-buf fd to the container
The container's GBM hands out buffers that are the display host's frames, and this is the phone end
of that: the server takes a frame from the host, offers its descriptor to the container over a unix
socket at a path both sides can reach, and takes the release back when the compositor is done with
it.

The channel is optional and silent by design. A compositor must render whether or not a container is
listening, and the render path must never wait on a socket - so it is opened lazily, every send is
bounded, and any error simply marks it down.
2026-10-01 03:37:50 +08:00
gold 1a7b4b6a2b [Feature] (gbm): the host-frame backend - the compositor's buffers are the phone's frames
The self-allocated backend makes the library stand on its own; this one makes it the thing the
project needs.  The display host on the phone owns a pool of frames, the render server draws into
them with Espryt and acks, and the host presents.  A gbm_bo created here is one of those frames: the
phone sends its dma-buf fd over a unix socket, the offer carries the geometry the host decided (not
the geometry the caller asked for), and destroying the bo sends the frame back.

Verified against a fake display host on this machine, with no phone involved: the buffer the fake
host fills is the buffer the bo maps - every pixel of the pattern is there - the fd the caller is
handed is a view of that same memory rather than a copy, a device with no frame waiting falls back
to allocating locally, and teardown closes exactly the descriptor it opened.  The self-allocated
path still passes its own test unchanged.

The channel is lazy and never blocking: a compositor must work whether or not the phone is there.
2026-10-01 03:27:22 +08:00
gold 55e5165e2f [Feature] (gbm): MobileGL's own GBM, so the compositor's buffers are the display host's frames
kwin never creates an EGL window surface.  Its undefined libgbm symbols are seventeen entry points
whose whole purpose is to hand out gbm_bo fds, and those fds are what Anland's display producer
sends to the daemon on the phone - so the question "whose pixels are on the screen" is answered by
"who allocated the bo", and today that is Mesa.  The client can be loaded, initialized and rendering
through glvnd with no window surface ever created, which is exactly what was measured.

This is a complete GBM: every public entry point of the vendored Mesa header, with the semantics a
compositor relies on - fds the caller owns, real mmap with refcounting, a swapchain that refuses to
hand out a busy buffer, import by fd, and teardown that leaves no descriptor or mapping behind.  The
buffers come from the display host's frames; where there is no host, the device allocates from the
dma-buf heap, falling back to an uncached heap and then to memfd, so the library stands on its own.

The vendored gbm.h is Mesa's, verbatim: the entry points are the ABI, not our idea of it.
2026-10-01 03:16:55 +08:00
gold 2860d09051 [Fix] (EGL): a second display is served, and glvnd gets its answers from one table
Two defects, both found by asking the client directly rather than by reading its log.

The backend refused to initialize any EGLDisplay but the first - "already bound to a different
EGLDisplay" - and the field it compared against has no reader anywhere else in the tree: the domain
rule was the whole of the rule.  What it cost was a compositor, which asks for the display it will
draw into AFTER probing another one.  The probe succeeded, the real display was refused, and the
caller reported the refusal one layer up as "no usable DRM render device" - a message that names
where it stopped, not why.

The second: glvnd asks a vendor for entry points from two places.  getProcAddress is how it builds
its own static EGL table, and getDispatchAddress is how it answers eglGetProcAddress for every name
it does not know itself - which is every extension entry point, the device queries among them.  This
side answered the second from a GL-only table, so an application asking for eglQueryDeviceStringEXT
got a null for a function the vendor implements.
2026-10-01 02:58:06 +08:00
gold c9206dd955 [Feature] (EGL): MobileGL serves EGL through libglvnd
On a system with libglvnd the libEGL.so.1 an application links is a dispatcher, not an
implementation, and the supported way to be the implementation it talks to is to be a vendor
library it loads and routes to.  On this machine it is also the ONLY way that reaches a caller which
does not link EGL directly: kwin goes through libepoxy, which dlopen()s libEGL.so.1 and dlsyms its
entry points at startup, so a library injected with LD_PRELOAD is invisible to it however loudly it
is named - measured, with the client in kwin's maps and not one call into it.

So this is the vendor half of that ABI: __egl_Main answers glvnd with the entry points by name, the
platforms this implementation serves, and the device it is associated with - the render node of the
machine, because that is what lets a compositor match its own node to the display it is about to
draw into, and because glvnd drops a vendor that does not answer at all.

Two details are the difference between a vendor that is loaded and a vendor that is used.  glvnd
requires setDispatchIndex to exist even when a vendor has nothing to record, which this one has -
its entry points are the implementation, they never re-enter glvnd.  And a vendor must hand over ITS
OWN entry points: with default visibility a reference inside this library resolves against the whole
process, where another libEGL that loaded first wins.  That is what -Bsymbolic-functions is for;
declaring the entry points protected says the same thing but does not compile against the platform
headers, which declare them default.
2026-10-01 02:58:06 +08:00
gold dfbda7ccb4 [Feature] (EGL): the platform and device extensions are one list, spelled once
A vendor string and an eglQueryString answer have to agree, and they are read by two different
parties: the system's libglvnd reads the first to decide which vendor a platform request should be
routed to, and an application reads the second before it decides to make that request at all.  Kept
apart they drift, and the drift has a signature - a library that is loaded, initialized, and never
called, while the caller reports that it found no provider for a function it never reached.

The list gains the device family for the same reason: a compositor asks which devices exist and
reads the DRM node off one before it asks for a display on it, so an implementation that answers
"no devices" is one it cannot use no matter how well the implementation draws.
2026-10-01 02:58:06 +08:00
gold 6a702afa72 [Chore] (glvnd): vendor the EGL vendor ABI headers
Two MIT headers from libglvnd, verbatim, because MobileGL is about to implement the vendor half of
glvnd's EGL ABI and the distribution ships the runtime without them.
2026-10-01 02:58:06 +08:00
gold ab685fb0e9 [Feature] (plugin): the server survives a dozing phone - a notification the user allows and a battery exemption the user grants
Two permissions decide whether a server on Android lives or dies, and both are invisible until
something goes wrong: without POST_NOTIFICATIONS (13+) a foreground service cannot show the
notification that is its reason to exist, and without the battery exemption a dozing phone freezes
the process - the listener stays in /proc/net and stops accepting, which from the other side of the
socket looks exactly like a network fault.

The control screen now shows the state of both and asks for whichever is missing, in the order that
makes sense: the notification prompt is a dialog of ours, the battery one leaves the app entirely,
and coming back from it re-runs the other branch through onResume. A denial is said out loud rather
than left to be discovered.

The manifest carries the two permissions; the service's own notification was already there.
2026-10-01 01:50:14 +08:00
gold 1ae7a8eaa4 [Feature] (disaggregated): WindowKind::HostFrame - the host-framed surface crosses the wire
The last thing between the host-framed surface and a run: the sub-byte on the wire that names it.
ServerOwned had one (7) and the host frame had none, so the encoder refused the op with
UnknownWindowKind - a correct refusal, and the reason the first run of MOBILEGL_IPC_SURFACE=host
failed in the codec rather than anywhere near the pixels.

 - protocol.fbs: WindowKind::HostFrame = 8, appended, legal on CreateWindowSurface only and always
   with nativeToken 0 - the same rules ServerOwned carries, for the same reason: no pointer of any
   process's window system crosses (Rule G/H).
 - SurfaceRefusal: HostFrameOnSetWindowHandle = 5, its own code rather than reusing the
   server-owned one: a refusal that names the wrong owner sends the next reader to the wrong place.
 - SurfaceOpCodec: the encode and decode arms, the two wire errors that go with them
   (HostFrameTokenNotZero latches as corruption, HostFrameOnSetWindowHandle is a named refusal), and
   the range bound that had ServerOwned as its upper limit now ends at HostFrame.
 - protocol_generated.h regenerated with the pinned submodule's flatc (scripts/gen_protocol.py),
   diff being exactly the two appended enumerators.

Both ends must be rebuilt together, as any protocol revision here: the container's client and the
device's server went out with this on either side.
2026-10-01 01:26:02 +08:00
gold ead7a10b68 [Feature] (disaggregated): the default framebuffer on a host-framed surface is the host's frame
Three things this path needs that the server-owned one never did, each because the pixels are not
in an EGL surface this time:

 - The client's DEFAULT framebuffer has to resolve to the frame.  Four places bind the driver's 0
   for the default framebuffer; they now ask g_hostFrameFramebufferId, which is 0 everywhere except
   while a host-framed surface is current, where it is the frame's framebuffer.  Without this every
   draw would land in the pbuffer the context was made with - a surface nobody ever sees.

 - eglMakeCurrent binds the surface's own default framebuffer behind the engine's binding shadow, so
   the frame has to be put back after every make-current, with the shadow invalidated first: a bind
   the shadow thinks already happened never reaches the driver.

 - The bind itself goes through FramebufferImpl::BindFramebufferId, not a raw glBindFramebuffer, for
   the same reason - a raw bind leaves the shadow claiming the previous framebuffer.

Frame lifecycle is otherwise unchanged: the frame becomes the draw target when it arrives, and is
dropped after the host has been answered and has presented it.

Compiles and links (libMobileGL.so, libMobileGLServer.so, exit 0), deployed to the device, and the
container's client is rebuilding against it.  The run that shows Plasma through it is next.
2026-10-01 01:14:51 +08:00
gold 035f1b07ae [Feature] (disaggregated): MOBILEGL_IPC_SURFACE=host - the display host owns the screen and the frames
The third on-screen shape, and the one this whole line was built toward: Anland keeps Wayland,
input and the display, and the render server draws into the frames the display host allocates.

    client (container)      asks for a window surface with no window of its own
    server (Android)        takes a frame from the host over the host frame bridge and binds it as
                            the draw target of that surface
    host (the app that
    owns the screen)        offers each frame, presents it when the server says it is drawn

The pieces, in the order a frame travels:

 - Config: MOBILEGL_IPC_SURFACE=host beside =server, and HostFrameWindowSurfaces() as the one
   predicate every client-side consumer asks (EGLImpl accepts a NULL window for either on-screen
   shape, and BackendObject_Remote routes the creation to the server instead of sending a handle).
 - Wire: the frame-local tag kHostFrameWindowBackend, the mirror of kServerOwnedWindowBackend - it
   names no WindowBackend of the pull build's enum, and the server consumes it before the unpack
   that would refuse it.
 - Server: ApplyHostFrameWindowSurface replaces nothing, unlike the server-owned arm: there is no
   window of either side's to substitute, because the backend takes the frames from the bridge.
 - Backend: SurfaceKind::HostFrame, InitHostFrameSurface (the EGL context comes from the pbuffer
   path; what changes is what it draws INTO) and PresentHostFrame (glFinish, answer the host, take
   the next frame) - and SwapEGLBuffers routes to it, because a host-framed surface has no
   swapchain to swap: the host is what puts the frame on the glass.

HostFrameTarget and HostFrameBridge, committed just before this, are what the backend uses.

Compiles and links in this tree (libMobileGL.so and libMobileGLServer.so, exit 0).  What is NOT
here yet, and is the next thing: the run that puts Plasma on the phone's screen through it.
2026-10-01 01:11:21 +08:00
gold c7ab618ab0 [Feature] (disaggregated): the host's frames as a draw target, and the socket they arrive on
Two modules, and nothing else changes yet: they are the half of the host-framed surface that is
about the frame itself, kept apart from the surface lifecycle so they can be reviewed and built
on their own.

HostFrameTarget (DirectGLES): an AHardwareBuffer becomes a renderbuffer through an EGLImage and a
framebuffer that is the drawing target while its frame is current.  The extension strings decide
first and the four entry points are taken from eglGetProcAddress only after they have, which is
the shape the T0 AHB import in Managers.cpp already uses; a driver that lacks one of them is
answered with its name, never an abort.  Every failure path drops what it made, so no GL name and
no EGLImage outlives a refusal.

HostFrameBridge (MG_Backend): the display host's own protocol, restated field for field - a
handle ahead of a fixed-size description on one AF_UNIX connection, then the answer, then the
host's readback of the pixels it can see.  The host dials; this side listens on the well-known
name.  A handle that arrives without its description is released before the failure is reported,
because a reference held past the connection starves the pool the host draws its frames from.

Why the host allocates the frames at all is measured, not chosen: AHardwareBuffer_createFromHandle
accepts an AHardwareBuffer this process allocated and refuses one the display queue handed over
(gralloc BAD_VALUE) - same process, same descriptor, same handle shape.  That result is what these
two modules are built around, and it is written down where the next reader will look for it.

Both compile against this tree's own flags.  Wiring them into the surface lifecycle, the swap path
and the client's opt-in is the next step, not this one.
2026-10-01 00:57:14 +08:00
BZLZHH 2e43df6125 [Docs] (disaggregated): the container as a client of the phone's on-screen server, and the two things that cost a run
Measured on the Redmi today: the Droidspaces container shares the phone's network namespace, so a
client in the container reaches the server's loopback listener with no adb forward and no LAN
address; the server owns the window (MOBILEGL_IPC_SURFACE=server), the Adreno does the drawing,
and a screenshot of the phone shows the colour that headless client asked for.

The two traps are written down because both cost a run: MOBILEGL_IPC_REQUIRE_SAME_BUILD=1 on the
server turns a Connect handshake's build-stamp warning into Refuse{BuildFingerprint} against a
release APK whose stamp is 0, and the client then dies on a caps gate that names something else.
2026-10-01 00:39:49 +08:00
BZLZHH 015b91f08d [Test] (extmem_probe): T4-host-image closes the round trip, and its first verdict was wrong for one constant
The leg now reports OK on the device: offered=3 imported=3 drawn=3 hostReadback=3 agreed=3 --
three self-allocated AHardwareBuffers from the display host imported as EGLImages by this side,
drawn into, and read back by the host at both the first pixel and the far corner.

Getting there took one measurement and one correction.  The handle message is 256 bytes on this
path and recvHandleFromUnixSocket drains it exactly (pending 256 -> 100, then 0 -> 100, then
0 -> 0), so the description that follows it was never the problem; the comparison was.  This leg
was checking the display host's messages against kT4Magic -- the dma-buf protocol's magic, which
belongs to the container peer -- while the host stamps AHB_BRIDGE_MAGIC on all three of its
messages.  A perfect round trip was reported as 'the offer's sideband did not follow its handle',
which named the wrong cause and cost the time.  kT4HostMagic is that value, and the leg now also
prints the handle's footprint and the bytes it actually read, so the next failure of this shape
cannot be mistaken for the other one.
2026-10-01 00:34:40 +08:00
BZLZHH 987b7ad13c [Test] (extmem_probe): T4-host-image is the other end - the buffer the display host is about to scan out, imported and drawn into, with the round trip closed by that host reading its own pixels back. 2026-09-30 23:53:49 +08:00
BZLZHH d26cf1e31b [Test] (extmem_probe): T4 asks the wrap question too - a driver without EGL_EXT_image_dma_buf_import is tried through AHardwareBuffer_createFromHandle before the offer is called untakeable, and both handle types are recorded so the answer names which one the platform reads. 2026-09-30 23:26:26 +08:00
BZLZHH 8196818e9f [Test] (extmem_probe): the T4 leg asks whether the render server can draw into an image allocated outside Android - and the device answers twice: this driver has no dma-buf import, Android-side allocation works end to end, and the offer must carry the plane layout rather than one linear plane. 2026-09-30 23:00:55 +08:00
swung0x48 648800a2da [Docs] (P8): close the phase - ruling ID-P8-19 (the exit device check and the P12 cross-host rerun), the P8 README and PLAN, the roadmap row and header, MEASUREMENTS row 19, the status page and the last debt 2026-09-30 08:25:38 -04:00
swung0x48 b4bee9dd38 [Docs] (P8): CONTRACT-P8 (the rules P8 landed, each with its code anchor and its test), rulings ID-P8-14..18 (MD + MF and the create-indirect fixture's alignment, SE, SV, the dev merge, the wave 2 integration gate) and the debts wave 2 hands on 2026-09-30 07:55:50 -04:00
swung0x48 494305381e [Merge] (P8): integrate wave 2 package MF - the push flush ladder gets dev's Adreno 830 queued-ring-copy guard (FlushPendingRangesFrom, create-instancing x Espryt x monolith 0.870 -> 0.99998 on the Redmi), and both G5 scripts pin the two reviewed ladder bodies under ID-P8-14 2026-09-30 07:12:50 -04:00
swung0x48 fb106a682e [Docs] (P8-MF): the package note - the push-ladder mirror of dev f973008c, create-instancing Espryt monolith 0.870 -> 0.99998 on the Redmi with spawn unchanged, why spawn was unaffected (the same two partial maps behind queued ring copies, the copy already submitted by the server's intervening whole-buffer maps), the red-onces and the gate 2026-09-30 07:09:28 -04:00
swung0x48 c8c4397475 [Merge] (P8): integrate wave 2 package SV - Magma's 3D native mip blit uses one layer, the depth/stencil mip decline gets its tally row, 1D depth mips decline by name, and the wire arm issues glDispatchComputeIndirect natively
# Conflicts:
#	MobileGL/MG_IntegrationTest/CMakeLists.txt
2026-09-30 06:11:04 -04:00
swung0x48 9c24a14d2b [CI] (scripts, P8-MF): pin G5's tenth row as well and re-pin both flush ladders on the reviewed post-fix bodies (ID-P8-14) - dev f973008c moved FlushPendingRangesNow past every base ref CI passes (the Adreno 830 queued ring copy) and ac9ab078 mirrors it into FlushPendingRangesFrom; both scripts' self-tests now run the pin-precedence and one-token controls on both ladders 2026-09-30 05:45:10 -04:00
swung0x48 ac9ab078cc [Fix] (Espryt, P8-MF): the push arm's flush ladder keeps a partial range-invalidating map behind an upload-ring copy still queued on the store, as dev f973008c's pull arm does - FlushPendingRangesFrom's ring tier stamps ringCopyRetireSerial and InvalidateFlushAccessFor refuses tier 1 for a partial range while that copy is unretired (the whole-buffer orphan-map stays exempt); Adreno 830 loses the queued copy and create-instancing on the monolith arm drew terrain sections empty 2026-09-30 05:43:55 -04:00
swung0x48 377d90d9f1 [Merge] (P8): integrate wave 2 package SE - the spawn/tcp default framebuffer publishes its surface to the client, the CPU mip arm declines on a driver-written base, D's case 3 runs on the Espryt split arms, the unmigrated-emulation name list is linted against its call sites, and the P8 notes' citations point at their content again 2026-09-30 05:24:11 -04:00
swung0x48 4cfb29a553 [Docs] (P8-SV): the note's gate table (G1 0/0, wire-declines 61/66, split_coverage, every lane, retrace 230/236) and the dirty-surface scan finding 2026-09-30 05:02:37 -04:00
swung0x48 51e95c9260 [Docs] (P8-SE): the note's gate and retrace readings 2026-09-30 04:47:27 -04:00
swung0x48 982958ebb1 [Test] (P8-SV): rename WireIndirectDispatchScenario's dispatch store member m_parameters -> m_groupCounts - gen_pipe_dirty_surface.py reads writes to shutter member names across all of MobileGL, and the name made NEW_PATCH_STATE's shutter look written outside its roots (self-test negative control 16 stopped tripping) 2026-09-30 04:42:06 -04:00
swung0x48 3717424229 [Merge] (P8): merge dev f973008c - the Espryt partial-flush fix for a queued ring copy on Adreno 830 (wave 2 MD), and dev's empty-binding uniform block fix deduplicated against P11's 7784d6bb (one code path), UnboundUniformBlockScenario registered on the split arms 2026-09-30 04:34:15 -04:00
swung0x48 f973008c4d [Fix] (Espryt): a partial buffer flush no longer takes the range-invalidating map while an upload-ring copy into the same store may still be queued - Adreno 830 loses the queued copy, and create-instancing drew whole Embeddium terrain sections empty 2026-09-30 04:07:41 -04:00
swung0x48 edcb9328e2 [Docs] (P8-SE): cite the integration CMakeLists by path - the bare basename resolves to the top-level file 2026-09-30 03:55:20 -04:00
swung0x48 e61060fbdf [Docs] (P8-SE): the package note, and the P8 package notes' line citations re-anchored by content at HEAD (A 14, B 29, C 10, D 12, E 18 plus the copy-image-shadow-mirror line, F 4) 2026-09-30 03:55:01 -04:00
swung0x48 3ce8ce3cb1 [Test] (P8-SE): the ForcedDs knob lanes run the default-framebuffer depth/stencil case again on all three arms (it was subtracted while red on spawn / tcp) 2026-09-30 03:55:01 -04:00
swung0x48 394d255487 [Docs] (P8-SV): the package note - the four items with their before/after, the design, the gates, ten red-once rows, the monolith observation and what was found but not fixed 2026-09-30 03:45:40 -04:00
swung0x48 0222852b94 [Fix] (Espryt, P8-SE): a spawn / tcp server publishes the default framebuffer's depth/stencil format (it has no frontend placeholder; the case leaves pending-fix for the three arms); the server CPU mip arm declines by name (generate-mipmap-store-declined) instead of filtering a store the driver wrote over, and E's refused-readback knob reaches it (CpuMipRefused. entries); WireIndirectDrawScenario case 3 runs on the Espryt split arms; copy-image-shadow-mirror goes and unmigrated_emulation_sites.py holds PipeCatalogueTest's list to the call sites 2026-09-30 03:44:29 -04:00
swung0x48 2edb2aa82a [Test] (P8-SV): WireMipShapeScenario (3D chain on both backends x every arm, 1D depth decline on Magma split+spawn, the 3D chain under the validation layer inproc) and WireIndirectDispatchScenario (both backends x every arm, Magma gated .IndirectDispatch.), the MipmapDepthStencilAspect tally in F1's depth/stencil decline case, WireDeclinePeek, and the two parity rows 2026-09-30 03:44:03 -04:00
swung0x48 fc415aee76 [Fix] (Magma, P8-SV): the wire arm's 3D native mip blit uses one layer with the depth in its z offsets, the depth/stencil mip decline gets its WireDeclineTally row (MipmapDepthStencilAspect), the depth shader mip refuses 1D/1D-array/3D view types before the chain grows, and glDispatchComputeIndirect is issued as vkCmdDispatchIndirect from the wire store (P8-D's barrier rule, wdsp counter, two named declines) 2026-09-30 03:44:03 -04:00
swung0x48 307267c188 [Docs] (P8): the monolith defect count is 12 (13 reports, one duplicate between C and D), 9 of them left to P13, not 14 and 11 2026-09-30 03:24:47 -04:00
swung0x48 cbbaf77f6d [Docs] (P8): wave 1 integrated - rulings ID-P8-5..13 (the survey, packages A-F, the integration gate, the wave 2 scope and the monolith defects left to P13), the survey note, open questions 9/11/15 answered, the debts and the P8 contract amendments, the P8 README table, PLAN-P8 wave 2 and the status page 2026-09-30 03:09:09 -04:00
swung0x48 39cd8fb73b [Merge] (P8): integrate package E - a texture readback the Espryt driver refuses is answered from the split server's staged store, which now follows glCopyImageSubData (FollowCopy, client-box merge, sticky driver-written mark, named decline texture-readback-store-declined); CopyImageStoreReadbackScenario on both backends x every arm plus the extent-scoped StoreRead knob; the dead CopyImageSubData+RENDERBUFFER latch goes (census 79 -> 78) 2026-09-30 02:17:51 -04:00
swung0x48 92eab9f3b7 [Merge] (P8): integrate package D - the Magma wire arm issues glDraw*Indirect / glMultiDraw*Indirect[Count] natively from the wire store (vkCmdDraw*Indirect[Count]) instead of a CPU expansion behind a GPU-idle wait, with an INDIRECT_COMMAND_READ barrier once per shader write; windr[] PipeStats counters and WireIndirectDrawScenario on both backends x every arm 2026-09-30 01:28:20 -04:00
swung0x48 9441071567 [Merge] (P8): integrate package C - the Espryt split server refreshes its staged copy of a buffer a shader wrote before a CPU reader consumes it (storage/atomic/buffer-image marks, XFB captures into the copy; restart rewrite, multi-draw rebase, *IndirectCount, CPU-fed indirect fields, XFB scatter and fp64 narrowing read through SplitHostBytesForCpuRead); GpuWrittenDrawInputScenario (12 cases) on the Espryt split arms 2026-09-30 00:45:16 -04:00
swung0x48 2f530ad8a2 [Merge] (P8): integrate package B - Espryt glGenerateMipmap on the server: the record's generated window (BASE/MAX clip, B1) and RGB16F/RGB32F natively or on the server's CPU from the driver's level (B2), retiring both split generate-mipmap names; GenerateMipmapServerScenario on both backends x monolith/Split/Spawn/Tcp plus the forced-CPU arm 2026-09-30 00:01:29 -04:00
swung0x48 c00f5eabae [Docs] (P8-E): the note's gate and retrace readings and two residuals (the extent-scoped tcp knob, the ERROR line before a store answer) 2026-09-29 23:51:49 -04:00
swung0x48 24b8aef616 [Fix] (P8 A): exempt F's Magma-only F1WireScenario.GenerateMipmapWithoutAShaderArmShapeDeclinesAndKeepsTheSession on DirectGLES as single-backend - it landed with P8-F after A's exemption table was cut, skips on Espryt like its F1Wire decline siblings, and split_coverage.py named it missing on the merged head 2026-09-29 23:41:10 -04:00
swung0x48 8133cd52ae [Merge] (P8): integrate package A - register the monolith-only cases on split/spawn/tcp (P8Coverage.cmake, five server-knob tcp supervisors), gate them with split_coverage.py and its exemption table, pin the tcp supervisor's Vulkan env 2026-09-29 23:37:12 -04:00
swung0x48 06f28b0a1b [Docs] (P8-A): the note says what the census measured about the audit's 396 and DoublePrecision on tcp, and the knob-port comment names the split/verify one-port overlap 2026-09-29 23:18:08 -04:00
swung0x48 b9feb9d7be [Docs] (P8-B): the package note - the generated window and the server CPU arm, the coverage matrix, seven red-onces and monolith defects M1-M3 recorded for dev 2026-09-29 23:16:05 -04:00
swung0x48 2b4ae539da [Merge] (P8): integrate package F - reclassify what our client cannot send: the multi-draw latches and Magma client-vertex-array become ProtocolCorruption, Magma mip shader-shape becomes a pre-growth named decline, unpublished cap bits and the index-mirror sentinel marked reserved 2026-09-29 23:07:03 -04:00
swung0x48 e815ce8d33 [Docs] (P8-C): the package note - readers and writers of the server's staged copy, the twelve GPU-written draw-input cases, each red-once, the retrace subset and the five monolith defects recorded for dev 2026-09-29 22:53:18 -04:00
swung0x48 00d6c2480e [Docs] (P8-E): the package note - the split read never touched a shadow, the store route and FollowCopy, the red-once table, the monolith defect for dev 2026-09-29 22:43:02 -04:00
swung0x48 f823422bfa [Fix] (Espryt, P8-E): a texture readback the driver refuses is answered from the server's staged store, which now follows glCopyImageSubData (FollowCopy, client-box merge, sticky driver-written mark, named decline); CopyImageStoreReadbackScenario on both backends x four arms plus the extent-scoped store-read knob; the dead CopyImageSubData+RENDERBUFFER latch goes (census 79 -> 78) 2026-09-29 22:43:02 -04:00
swung0x48 e52a997f34 [Fix] (Espryt, P8-C): the split server refreshes its staged copy of a buffer a shader wrote before a CPU reader consumes it - storage/atomic/buffer-image writes mark the server twin, XFB captures land in the copy, the restart rewrite, multi-draw rebase, *IndirectCount, CPU-fed indirect fields, XFB scatter and fp64 narrowing read through SplitHostBytesForCpuRead; GpuWrittenDrawInputScenario (12 cases) on the split arms 2026-09-29 22:38:33 -04:00
swung0x48 c7a14ed790 [Test] (P8-F): give the mip shader-shape decline its own .ShaderMipDecline. tail - spawn_lane_parity.py holds each server-knob tail to exactly one case, and sharing .ShaderMip1. named two 2026-09-29 22:37:49 -04:00
swung0x48 d639e54616 [Perf] (P8-D, Magma): the wire arm issues glDraw*Indirect / glMultiDraw*Indirect[Count] as vkCmdDraw*Indirect[Count] from the wire store instead of expanding them on the CPU behind a GPU-idle wait (create-indirect split: 321 waits / ~8 s -> 0), with an INDIRECT_COMMAND_READ barrier once per shader write; windr[] PipeStats counters, WireIndirectDrawScenario on both backends x every arm, notes/p8/D.md 2026-09-29 22:32:49 -04:00
swung0x48 32f222dbc2 [Docs] (P8-A): the package note - the 304-case census per arm, the server-knob lanes and their supervisors, the exemption classes, the red-once record, lane wall times and the findings 2026-09-29 22:30:20 -04:00
swung0x48 a2c81a2f64 [Fix] (P8-A): the tcp lane's shared supervisor carries the Vulkan ICD and the three lavapipe iterationRP repairs, which Magma reads in the server - DirectVulkan.Tcp IterationRPProgram203 missed its golden output by one texel without them 2026-09-29 22:30:20 -04:00
swung0x48 8dac1fd8a4 [Fix] (Espryt, P8-B): glGenerateMipmap on the server - the record's plan is the generated window (B1) and RGB16F/RGB32F generate natively or on the server's CPU from its own level (B2), retiring both split generate-mipmap names; GenerateMipmapServerScenario on both backends x monolith/Split/Spawn/Tcp plus the forced-CPU arm 2026-09-29 22:22:55 -04:00
swung0x48 b71cb9cc2c [Docs] (P8-F): cite MG_Backend/Init.cpp by path - the bare basename is ambiguous 2026-09-29 22:15:45 -04:00
swung0x48 efd3533bd1 [Docs] (P8-F): the package note - the four items, their evidence, the red-once record and what was found but not fixed 2026-09-29 22:15:29 -04:00
swung0x48 67890f523a [Refactor] (P8-F): reclassify what our client cannot send - the multi-draw client-indices/instanced latches and Magma's client-vertex-array become ProtocolCorruption, Magma's mipmap-shader-format-or-shape becomes a named decline decided before the chain grows (with a knob-forced 3D case on split and spawn), and the three unpublished cap bits and the index-mirror sentinel are marked reserved 2026-09-29 22:15:29 -04:00
swung0x48 976fbe4cd4 [Test] (P8-A): register the monolith-only integration cases on the split, spawn and tcp arms (P8Coverage.cmake: 246 ambient cases, five server-knob lanes each on its own tcp supervisor), retire the ClientSideIndices and GuiBatch exclusions, and gate the rest with split_coverage.py and its exemption table 2026-09-29 22:14:31 -04:00
swung0x48 8d3e2317a6 [Docs] (P8): re-scope the phase - strike the original scope in the roadmap and the P8 README, add PLAN-P8 (six items closed, packages P8-0 and A-G, the rewritten gates) and rulings ID-P8-1..4, and mark P8 in progress on the status page 2026-09-29 21:31:15 -04:00
swung0x48 0b55b55d4d [Docs] (P11): the status page lists the decisions waiting on the user - the P8 start and the six debt rulings 2026-09-29 21:17:21 -04:00
swung0x48 642ac891af [Docs] (P11): ruling ID-P11-20 - the closing head's gate, lanes and retrace numbers 2026-09-29 21:13:12 -04:00
swung0x48 592558f5fc [Docs] (Debts): the debt review - drop the closed E1, DualBlock and GetCaps-blobref rows (open question 24 closed, the blobref rule now a CONTRACT-P5 row-18 invariant), move the unmeasured M2 note to MEASUREMENTS, rewrite HyperOS / P11 RSS / named refusals, split the P12 row, merge the two CI rows, add three B2 rows, and mark each destination as needing a ruling or ready 2026-09-29 21:12:28 -04:00
swung0x48 0a117243d5 [Docs] (P11): close the phase - rulings ID-P11-17..19 (B2 T0 zero-copy, the stale indexed binding and Magma empty-uniform fixes with the skip-hole, T0 on by default), B2-DEVICE, the P11 README, the roadmap row and header, the status page and the debts P11 hands on 2026-09-29 20:33:00 -04:00
swung0x48 344e2f8439 [Docs] (P11 B2): T0 is the default (the user, 2026-09-29) - CONTRACT-P11 B2's tier table, unset versus 0, the revision-5 fix-up; the knob rows in design/09 2026-09-29 20:15:29 -04:00
swung0x48 91900bdc5e [Test] (P11 B2): the unset knob on a stream and on a no-AHardwareBuffer shared-segment session - T2, no W/I fallback line, the fallback counted; the 0x80 ask accepted and 0x81 refused 2026-09-29 20:15:29 -04:00
swung0x48 b32140e614 [Feat] (MG_Remote, P11 B2): T0 is the default - an unset MOBILEGL_IPC_ADOPT_TIER asks T0 (ask 0x80) and falls back to T2 quietly (MGLOG_D, still counted: client T0Fallbacks, server T0Refusals), while an explicit 0 keeps every named line; revision 5's schema comment and pin fixed up 2026-09-29 20:15:29 -04:00
swung0x48 7030d307be [Test] (Magma): UnboundUniformBlockScenario - a dispatch and a draw whose program declares a uniform block at an empty binding point (never bound, or its buffer deleted while bound) run and produce their fallback, and a buffer bound there afterwards is read, on both backends 2026-09-29 19:58:41 -04:00
swung0x48 853c5f1264 [Fix] (Magma): a uniform block declared at an empty binding point reads a zeroed block instead of dereferencing a null buffer 2026-09-29 19:55:53 -04:00
swung0x48 e44d6ac7da [Docs] (P11 B2): CONTRACT-P11 - a buffer deleted while bound to an indexed point: the defect, the fix, Magma's behaviour before it, the tests with their red-once, and the G1 delta 2026-09-29 19:44:08 -04:00
swung0x48 8571e0c001 [Test] (P11 B2): DeletedBoundBufferScenario - a buffer deleted while bound to a storage, uniform, atomic-counter or transform-feedback point, its slot reused, then a dispatch and a draw, on monolith and Split/Spawn/Tcp for both backends; and the one-process arena pair as a regression 2026-09-29 19:32:19 -04:00
swung0x48 7784d6bbf7 [Fix] (Magma): a uniform block declared at an empty binding point reads a zeroed block on the monolith arm instead of dereferencing a null buffer, as the wire arm already does 2026-09-29 19:32:19 -04:00
swung0x48 4aec5621a8 [Fix] (MG_State, P11 B2): glDeleteBuffers moves the bind-point generation of every indexed target it unbinds, and a transform-feedback object swap moves it too - the server's window no longer names a deleted buffer's handle after its slot is reused 2026-09-29 19:32:19 -04:00
swung0x48 4c02279f22 [Docs] (P11 B2): CONTRACT-P11 B2 - the Espryt split/spawn itest entries now pin MOBILEGL_ITEST_REQUIRE_GPU themselves 2026-09-29 19:12:46 -04:00
swung0x48 e869d18ce8 [Test] (P11 B2): pin MOBILEGL_ITEST_REQUIRE_GPU=1 on every Espryt split/spawn itest environment, as the Magma and Espryt TCP lists do - a handshake-time Fatal fails there instead of skipping as a ctest pass 2026-09-29 18:57:04 -04:00
swung0x48 9d1e788597 [Docs] (P11 B2): CONTRACT-P11 B2 - M2 staying green on Espryt is consistent with the driver holding its own reference, not verified 2026-09-29 18:44:25 -04:00
swung0x48 ce5753c883 [Docs] (P11 B2): CONTRACT-P11 B2 - the tier table that replaces A1's shared-segment Fatal, revision 5, the Offer and its thread, POST, the staged-copy audit, lifetime, CPU reads, host and device gates with their red-once 2026-09-29 18:24:53 -04:00
swung0x48 5e65907d63 [Test] (P11 B2): RemoteClientAdoptT0.KnobTwoIsTodaysT2WithNoT0Line - the default asks 2 and neither side settles, imports or logs T0 even where the stand-in platform could run it 2026-09-29 18:20:25 -04:00
swung0x48 98bf1925aa [Test] (P11 B2): LargeArenaAdoptionScenario - a draw still queued when the arena is rewritten or respecified keeps the bytes it was issued against (no readback between the two draws, so an in-place write or an early free shows) 2026-09-29 18:06:48 -04:00
swung0x48 17f1c0b59b [Tool] (P11 B2): extmem probe --hold-ms - the T0S child keeps its AHardwareBuffer locked while the parent holds both imports, for the dmabuf memory calibration 2026-09-29 17:50:50 -04:00
swung0x48 2522f1803c [Test] (P11 B2): T0 protocol on the host - the Offer inbox (late, out of order, stale, missing) over both aux channels, the whole T0 path on a stand-in platform, A1's two handshake-Fatal cases rewritten as named fallbacks, revision 4 refused by name, AdoptTier0 lanes on the shared-segment arms and the allow switch through spawn 2026-09-29 17:50:50 -04:00
swung0x48 7cde55b1c1 [Feat] (Magma, P11 B2): import a T0 store through VK_ANDROID_external_memory_android_hardware_buffer as a wire buffer (external VkBufferObject), serial-gated release, done-no-bytes readback, and the sustained-lock POST on the renderer's device 2026-09-29 17:50:50 -04:00
swung0x48 2a69f61c88 [Feat] (Espryt, P11 B2): import a T0 store with glBufferStorageExternalEXT as the adopted twin - no staged copy, server writes queued GPU-ordered, CPU readers on the import's coherent map, fence-gated release, re-import after a context loss, and the sustained-lock POST in a throwaway context 2026-09-29 17:50:50 -04:00
swung0x48 590d74525d [Feat] (MG_Remote, P11 B2): T0 behind MOBILEGL_IPC_ADOPT_TIER - control revision 5 (LinkTerms.adoptTier ask), kCapAdoptT0 granted at the first native bind after the backend's POST, the AHardwareBuffer Offer on the aux socket matched to its map_persistent record by seq, named T2 fallbacks instead of A1's handshake Fatal, the server allow switch, done-no-bytes CPU reads 2026-09-29 17:40:56 -04:00
swung0x48 b57276b88a [Tool] (extmem_probe, P11 B2): route mode - the probe serves in the render server's place (-Pmobilegl.extmemProbeAsServer) and --route-client sends a held-lock AHB over B1's fd: pair, directly on aux and through a socketpair hop 2026-09-29 16:44:37 -04:00
swung0x48 ac25e3323a [Tool] (extmem_probe): build inside the API-26 trace APK - memfd_create through syscall() below API 30, the two Vulkan 1.1 calls through vkGetInstanceProcAddr below API 28 2026-09-29 16:31:52 -04:00
swung0x48 96881f639c [Build] (P11 A3): MOBILEGL_BUILD_EXTMEM_PROBE / -Pmobilegl.buildExtmemProbe packages tools/spikes/extmem_probe as libMobileGLExtmemProbe.so in the trace flavour 2026-09-29 16:31:52 -04:00
swung0x48 16ea23c927 [Tool] (extmem_probe): marker/args-file mode for the trace APK exec hook, --regions-at-end, and the T0S sustained-lock rows (Vulkan copy/fill + GL compute on a never-unlocked client AHB) 2026-09-29 16:31:52 -04:00
swung0x48 b9b14c4775 [Docs] (P11): status page - the two handed-on dev bugs are merged, the gate row points at the B1 and M2 rulings 2026-09-29 16:14:34 -04:00
swung0x48 37c7aede31 [Docs] (P11): ruling ID-P11-16 on B1 (fd: endpoint, token broker pairing at hand-off, app_process helper, per-peer apply affinity, the HyperOS battery requirement, the device pass), the B1 debts, the status page, the phase README and the roadmap row 2026-09-29 16:14:18 -04:00
swung0x48 64b5794bc9 [Docs] (P11 B1): CONTRACT-P11 B1 - under the EOF Hello wait a second client slower than 2 s sees a send failure rather than Refuse{Busy}, and the unix-reach argument is the device's 2026-09-29 16:10:45 -04:00
swung0x48 ce4dca2835 [Docs] (P11 B1): CONTRACT-P11 B1 - the rebased head's device pass: rd12 shm vs tcp on both routes, Magma openra re-measured clean, the 10 s slow start and the plugin flavour 2026-09-29 16:06:30 -04:00
swung0x48 df826f426f [Docs] (P11 B1): CONTRACT-P11 B1 - the broker pairs at hand-off, a paired unix session waits for its Hello until EOF (the decision and its file:line), the lone-prime guard, the supervisor's announced core as the one source, and the follow-up gates 2026-09-29 15:03:52 -04:00
swung0x48 7ccb920ad2 [Feat] (Plugin, P11 B1): the broker presents the PairBind pair at hand-off (PairBindFrames, the encoder's bytes) and answers v2 with that flag and the apply core its supervisor announced; the helper passes MOBILEGL_IPC_FD_PAIRED=1 on, and Java no longer re-derives the reserved core 2026-09-29 15:03:52 -04:00
swung0x48 582adb0dd6 [Test] (P11 B1): the broker's PairBind templates against the encoder, a pair presented at hand-off whose Hello comes 5 s later, a pair presented only at program start refused by name, and the reserved core on lone-prime, symmetric and unreadable topologies 2026-09-29 15:00:30 -04:00
swung0x48 578795c84d [Feat] (MG_Remote, P11 B1): pairing at hand-off - the fd: endpoint presents no PairBind when MOBILEGL_IPC_FD_PAIRED=1 says the broker did, names the server's refusal when its own PairBinds arrive too late, and a paired unix session waits for its Hello until EOF instead of 10 s 2026-09-29 15:00:30 -04:00
swung0x48 9af4c8aefd [Feat] (MG_Remote, P11 B1): the reserved apply core never takes a lone prime core - several prime cores give the lowest, one gives a big core beside it or none, and the log names which - and the supervisor announces the core it reserves on stdout at start-up, the one source the broker relays 2026-09-29 15:00:30 -04:00
swung0x48 5a97eb8d1b [Docs] (P11 B1): CONTRACT-P11 B1 - anchor the evidence paths to ~/w7/notes/p11/evidence/b1 and the report 2026-09-29 14:47:31 -04:00
swung0x48 fc427a14d6 [Docs] (P11 B1): CONTRACT-P11 - the external-client route (helper, token broker, fd: endpoint), the token rule and the named refusals, the apply-thread policy by peer and plane with its selection measurements, the keep-alive setting users must make, the ROM dependence and the gates 2026-09-29 14:47:31 -04:00
swung0x48 54e7c92cf4 [Feat] (Plugin, P11 B1): the broker hands the helper the core the server reserves for the session's apply thread and the helper execs the program through taskset on every other core (--keep-affinity opts out), so a client the scheduler would put there is not starved 2026-09-29 14:47:31 -04:00
swung0x48 1abddf0d00 [Test] (P11 B1): ApplyThreadPolicy - every row of the auto table by peer and plane, the Redmi topology, the shapes with no prime pair to split, explicit masks, spins and unrecognised text 2026-09-29 14:47:31 -04:00
swung0x48 0819f9d410 [Feat] (MG_Remote, P11 B1): the apply thread's auto affinity and spin chosen per session from the peer's dial mode and data plane - a dialled-in shared-segment client leaves the apply thread only the lowest prime core, stream, forked and in-process peers keep every big core 2026-09-29 14:47:31 -04:00
swung0x48 952a1ca4a7 [Tool] (P11 B1): trace flavour gate tooling - package the retrace CLI as libMobileGLRetrace.so (packageRetraceExe) and run an external client as a child of the app kept in front (mobilegl_exec_argv) 2026-09-29 14:47:31 -04:00
swung0x48 bd05e061fc [Feat] (Plugin, P11 B1): same-device external clients - the exported token-checked broker that connects twice to the server's own endpoint and answers every request, the app_process helper that execs the program on the pair, a private broker listener beside a TCP-only supervisor, and the copyable launch command on the server screen 2026-09-29 14:47:31 -04:00
swung0x48 23a9470a28 [Test] (P11 B1): the fd: endpoint end to end against a real server process after two bare connects, its PairBind pair read off the server side, and its named refusals that write nothing 2026-09-29 14:47:31 -04:00
swung0x48 53098bc57f [Feat] (MG_Remote, P11 B1): the fd:<control>,<aux> client endpoint - adopt two connected unix sockets a broker opened for this process, refuse by name anything that is not a distinct connected AF_UNIX stream pair, and present the client's own PairBind pair before the Hello, over shared segments 2026-09-29 14:47:31 -04:00
swung0x48 76f75f3a0c [Docs] (P11): ruling ID-P11-15 on the M2 dev merge - the scattered-upload fix completed on the split arms, the trim fence re-audited on the split server, the census and ratchet notes; DEBTS closes the two merged dev bugs and takes M2 leftovers 2026-09-29 13:46:37 -04:00
swung0x48 d587d394fe [Docs] (P11): zero-copy stays switchable (ruling ID-P11-14) - MOBILEGL_IPC_ADOPT_TIER 2 (default) keeps today shared-memory push, 0 takes the T0 import, and an unusable T0 falls back to T2 with a named line; the server gets its own allow switch 2026-09-29 13:45:57 -04:00
swung0x48 de996d539f [Test] (P11 M2): DescriptorPoolGrowthScenario waits for the inproc server to apply every record it emitted before each census read - under run-ahead a census right after the burst's swap read the pools before the burst, or a trim a frame late 2026-09-29 13:29:38 -04:00
swung0x48 b4b1825951 [Docs] (P11 M2): TextureEmit.h and the applier's accumulation name the one box-only record that is not a whole level - the oversize fallback's withheld cover, a partial box spanning gaps 2026-09-29 12:36:41 -04:00
swung0x48 ec5f6ead3f [Test] (P11 M2): TextureEmit.MoreThanKMaxDirtyRectsCollapsesToTheBoxWithRegionCountZero keeps its name and pins the one collapse left - a footprint larger than a record carries falls back to the storage's rect list 2026-09-29 12:26:10 -04:00
swung0x48 f636a4cdae [Test] (P11 M2): the link ratchet records TextureObjectWithOneMipmap::GetStorageDirtyFootprint - dev 9524c688's new virtual on the placeholder vtable UniformManager already reaches (171 -> 172) 2026-09-29 12:26:10 -04:00
swung0x48 bd2e43ede3 [Test] (P11 M2): dev's two Magma scenarios on the split arms - both on Espryt's three arms, OversizedRecordingSplitScenario on Magma's three (and pinned to a 64-draw budget on split and spawn, where the server reads it), DescriptorPoolGrowthScenario's census cases on Magma's split arm only; the split-recording arming reads both role logs and dev's pinned lane holds its log's RESOURCE_LOCK, and the census cases count the sets the wire arm holds between its descriptor rewinds 2026-09-29 12:20:15 -04:00
swung0x48 e2333d98a8 [Fix] (Magma, P11 M2): DirectVulkan::GetDescriptorPoolCensus reads the server renderer's pools on the apply thread under inproc (the test probe, between drain batches) instead of racing its growth, rewinds and trims, and answers unavailable on spawn and tcp 2026-09-29 12:20:15 -04:00
swung0x48 5c9ad5b499 [Test] (P11 M2): RemoteGuards pins GetStorageDirtyFootprint on the apply thread as Fatal{RoleViolation, "texture-legacy-arm"}, beside the rest of the dirty model's rows 2026-09-29 12:16:55 -04:00
swung0x48 9145433890 [Fix] (P11 M2, Espryt, Magma): the texture emitter carries a level's exact dirty footprint as its regions (dev 9524c688) - a single partial box as one region, only a whole-level write as none - so both servers upload exactly the texels written over GPU content; the applier keeps it exact up to 4096 regions, and Espryt's handle arm applies the storage's box-or-rects cut-offs itself while its texture holds no GPU writes 2026-09-29 12:16:55 -04:00
swung0x48 c371599811 [Merge] (P11): dev@c14283ee - scattered sub-image writes upload exactly their texels, Magma submits a frame's recording every 16384 draws with backpressure, and Magma grows descriptor pools geometrically and trims a quiet slot whole; the handle arm notes driver-side writes by handle and the wire arm splits recordings and keeps the trim's epoch peak across descriptor rewinds 2026-09-29 12:11:37 -04:00
swung0x48 b34eed580d [Docs] (P11): HSPIKE - an app_process helper with no Context gets connected sockets to the server app over Binder under the shell uid and another app uid, token-checked; ruling ID-P11-13 on the single broker mechanism and the ROM caveat 2026-09-29 11:42:59 -04:00
swung0x48 c14283ee2d [Docs] (Magma): the trim threshold's comment states the bound its argument proves - a growth adds at most twice what the slot held, so the slot stays over a third full 2026-09-29 10:58:33 -04:00
swung0x48 9dccabf03c [Test] (Magma): descriptor pool growth and trim - the pure sizing policy adds O(log N) pools for an N-set frame, and DescriptorPoolGrowthScenario's 4096-distinct-set frame lands whole on both backends, grows at most 7 pools, and is given back after quiet frames and re-grown while they are still in flight 2026-09-29 10:34:16 -04:00
swung0x48 53fa16c2c2 [Fix] (Magma): a frame slot whose descriptor pools stay under a quarter full for 120 frames is handed back whole - its set cache cleared, grown pools destroyed, base pool reset and every memo holding a set handle dropped - so rd12's loading burst no longer lives for the whole process (live sets 15,042 -> 39, pools 65 -> 6); MOBILEGL_MAGMA_DESCRIPTOR_TRIM_FRAMES (0 = never) and a test-visible pool census 2026-09-29 10:34:16 -04:00
swung0x48 c2de6576ad [Fix] (Magma): a frame slot's next descriptor pool doubles its LARGEST pool of the same update-after-bind flavour, capped at 4096 sets - doubling the first (oldest, smallest) one made every grown pool the same size, so rd12's 15,016-set loading frame grew one slot to 60 pools (1x128 + 59x256); now 9 (64 ... 4096, 4096, 4096) 2026-09-29 10:34:15 -04:00
swung0x48 35e1c861a3 [Docs] (P11): ruling ID-P11-12 on pairing by PairBind nonce and revision 3 -> 4; B0 F2 corrected - the product TCP path already paired by identity, the defect was unix endpoints and the one-shot TCP server; the TCP fixture comment no longer names AcceptPair 2026-09-29 10:20:53 -04:00
swung0x48 9ebcf5a73d [Docs] (P11 PAIR): CONTRACT-P11 - the PairBind wire change and revision 3 -> 4, the acceptor's rules, cross-revision behaviour, the gates and the red-once 2026-09-29 10:19:59 -04:00
swung0x48 05dc94fa38 [Test] (P11 PAIR): PairAcceptorTest - a stray before a client, four interleaved connections, an orphaned aux past the budget and a Hello-first client, each on unix and tcp - and the device's probe-then-client case end to end in ServerSpawnTest 2026-09-29 10:19:59 -04:00
swung0x48 abe29eae2d [Fix] (MG_Remote, P11 PAIR): a socket client's control and aux connections pair by the nonce each presents first, not by arrival order - a readiness probe or a second client can no longer take half a session, and a client that sends its Hello first is refused by name 2026-09-29 10:19:59 -04:00
swung0x48 0a6af63655 [Feat] (P11 PAIR): the PairBind control message - the client's nonce and which of its two connections this is, the first frame on each - with the control revision 3 -> 4 and its pin 2026-09-29 10:19:59 -04:00
swung0x48 32b81dae42 [Docs] (P11): the two dev bugs P11 handed on are fixed on dev (Espryt scattered uploads 9524c688; Magma rd12 command-buffer growth d20c2d01) and arrive with the next dev merge 2026-09-29 09:52:37 -04:00
swung0x48 9c32d9fa68 [Test] (Magma): OversizedRecordingSplitScenario carries state across the split - a depth occluder, a stencil mark, an additive blend destination, a 4x multisample target resolved after it, occlusion and primitives-generated queries and a transform-feedback capture left open across it, each checked on both backends and in the budget-64 lane 2026-09-29 09:44:58 -04:00
swung0x48 4d3d677a7f [Test] (Magma): OversizedRecordingSplitScenario - a 1024-draw frame, one pixel and one uniform value per draw, lands whole on both backends; the DirectVulkan.SplitRecording lane pins the draw budget to 64 and asserts the split fired 2026-09-29 09:44:58 -04:00
swung0x48 d20c2d01ae [Fix] (Magma): a frame's recording is submitted every 16384 draws/dispatches and continues on a fresh command buffer, with at most frames-in-flight + 1 submissions outstanding - rd12's 981k-draw loading frame no longer grows one command buffer until Adreno's 16 KiB driver mappings exhaust vm.max_map_count (MOBILEGL_MAGMA_MAX_DRAWS_PER_COMMAND_BUFFER, 0 = unbounded) 2026-09-29 09:44:58 -04:00
swung0x48 2fcde6eb9b [Docs] (P11): B0-CROSS-APP - SELinux refuses a foreign context the connect to the server app unix endpoint, past a brokered connect the shared segments work unchanged, default placement makes shm slower than tcp; ruling ID-P11-11 on what B1 must build, and the HyperOS uid freeze as a debt 2026-09-29 09:34:26 -04:00
swung0x48 21c7e8cd63 [Test] (MG_IntegrationTest, MG_Test): scattered writes over a GPU clear on both backends - two distant 2x2 boxes, a hundred boxes, boxes filling 3/4 of their union and three quadrants spanning the level, the 3/4 and level-spanning cases again at the image-bindable re-mint, and a MipmapStorage footprint fuzz against a reference bitmap 2026-09-29 09:06:07 -04:00
swung0x48 9524c688ca [Fix] (Espryt, Magma): scattered glTexSubImage writes upload exactly their texels - MipmapStorage keeps an exact dirty footprint beside its rect list (a write log once a merge or a fold would widen the list past the writes), Magma always uploads it, Espryt uploads it box by box (through the unpack ring too) once the driver texture may hold GPU writes and merges it at the image-bindable re-mint; the union box, the 3/4 cut-off, the 96-rect fold and a level-spanning union no longer put the stale shadow over GPU-written texels (two 2x2 writes over a GPU clear: 41 of 64 texels red -> 0) 2026-09-29 09:06:07 -04:00
swung0x48 6481691d1e [Docs] (P11): the status page, the roadmap row and the phase README for P11 in progress, and ruling ID-P11-10 on absorbing origin@86dc38a0 2026-09-29 08:06:38 -04:00
swung0x48 9176dcff3d [Merge] (P11): origin feat/disaggregated@86dc38a0 - its eglSwapInterval port is dev c60639db (already merged) plus ForgetRequestedSwapInterval, so an ended Magma server session no longer hands its interval to the next 2026-09-29 08:05:57 -04:00
swung0x48 26de2b60fc [Docs] (P11): rulings ID-P11-1..9 with the integration gate, A-DEVICE (T0 is GO from the app domain on both backends; adoption buys RSS not frame time on the Adreno; split T2 costs RSS and p99), the debts the dev merge closes and the ones P11 hands on, and W2-REMINT marking the monolith re-mint fixed 2026-09-29 08:03:09 -04:00
swung0x48 82467823ef [Docs] (P11 C): the contract's A1/A2 sections, Config.h's GPU-ordered landing, Init.cpp's DirectVulkan resource table, and the stream refusal CONTRACT-P6 / design 06 / 07 claimed now cites its code 2026-09-29 07:23:41 -04:00
swung0x48 2a902cbd79 [Test] (P11 A1/A2): LargeArenaAdoptionScenario asserts the tier its arena landed in and reads the tcp AdoptTier0 refusal; registered on Split/Spawn/Tcp for both backends (Espryt through the split-arms macro, Magma's Arena. block and Buffers lane), the monolith lanes declare adopted 2026-09-29 07:23:41 -04:00
swung0x48 3c72e5c637 [Fix] (P11 A1): settle MOBILEGL_IPC_ADOPT_TIER at the handshake from the data plane - a stream refuses T0/T1 by name and runs T2 on either side, a client over shared segments dies by name before any record, the at-use check trusts the settlement; unit gates on a real tcp session and a real inproc handshake 2026-09-29 07:23:41 -04:00
swung0x48 1797db0070 [Docs] (P11): B0 drives the cross-app client first as a retrace executable run from adb shell, then re-checks the connection and fd passing from a retrace client inside an app, the domain a Termux-style client really runs in 2026-09-29 07:23:41 -04:00
swung0x48 01a33dfc58 [Docs] (P11): B0 tests the cross-app client as a retrace client in one app against a render server in another - no Termux route; the fallback channel is a bound Service in the server app handing over a socketpair end 2026-09-29 07:23:41 -04:00
swung0x48 93a72678ca [Docs] (P11): B is scheduled - the consumer is a GL program started from another app on the same device (Termux and the like); B0 probes a cross-app fd channel, B1 moves such a client onto shared-memory segments, B2 adds the T0 import 2026-09-29 07:23:41 -04:00
swung0x48 3cb3e32b87 [Test] (P11 M): ImageBindableRemintScenario pins a barrier before its imageLoad copy - the kept-allocation case read the pre-clear upload once on push-monolith under load, the ID-P9-11 shape 2026-09-29 06:47:11 -04:00
swung0x48 89f1e6cfba [Test] (P11 M): P4aFinalFixScenario expects no re-mint count on the monolith arm either - dev's 568090f0 keeps an immutable, unwidened allocation there as the server already did 2026-09-29 06:47:10 -04:00
swung0x48 2c7d99d5e8 [Test] (P11 M): the server half of a split eglSwapInterval - Magma's backend object carries the request into its table's SetSwapInterval (null before dev@c60639db) 2026-09-29 06:33:36 -04:00
swung0x48 9cde08064c [Test] (P11 M): dev's ImageBindableRemint and PartialUploadOverGpuContent scenarios on the Split/Spawn/Tcp arms of both backends, three scattered-upload cases for the handle arm's carried layout, and TextureRemintPullScenario's monolith Espryt skip retired 2026-09-29 06:30:54 -04:00
swung0x48 48a306025d [Fix] (Espryt, P11 M): the push builds' monolith re-mint merges the applier's pending boxes over the GPU readback - the client clears its dirty flags at emission there, so a pending glTexSubImage was read back over and lost 2026-09-29 06:30:54 -04:00
swung0x48 e0102bad8c [Merge] (P11): dev@c60639db - the Espryt image-bindable re-mint reads GPU levels back on the monolith arm too, partial writes into converted or reshaped levels upload only their region on every arm, Magma orders an upload after the recording's writes and queued clears, and Magma's eglSwapInterval reaches the swapchain 2026-09-29 06:15:56 -04:00
swung0x48 ffeed0da3a [Docs] (P11): the re-scoped plan - each roadmap item against the tree at 74ae652e; A makes the adopt tier honest (client-side verdict, the Stream refusal, a tier assertion, three arms) and re-runs open question 3 and the baseline on the device, B (T0) is conditional, C closes by docs 2026-09-29 05:52:51 -04:00
swung0x48 86dc38a07e [Fix] (Magma): eglSwapInterval reaches the swapchain - 0 picks IMMEDIATE, else MAILBOX, else FIFO; >= 1 FIFO; < 0 FIFO_RELAXED, else FIFO; rebuilt after the next present when the mode changes, and an ended server session forgets its interval (MC 26.3 inproc vsync on: flat 120) 2026-09-29 05:52:22 -04:00
swung0x48 c60639db1d [Fix] (Magma): eglSwapInterval reaches the swapchain - 0 picks IMMEDIATE, else MAILBOX, else FIFO; >= 1 FIFO; < 0 FIFO_RELAXED, else FIFO; rebuilt after the next present when the mode changes (MC 26.3 vsync on: 190-290 fps -> 120) 2026-09-29 05:51:18 -04:00
swung0x48 74ae652ecd [Docs] (P10): close the phase - CONTRACT-P10 sections 2-3, rulings ID-P10-1..14 with the integration gate, C-MEASUREMENTS (device fence bench, loopback credit 1/2/3, the retrace round-trip census), the roadmap row, the debts P10 hands on, design/09, the phase README and the status page 2026-09-29 05:29:00 -04:00
swung0x48 414dd09bda [Test] (P10): the fence bench comment - the replay corpus does carry timed fence waits (improved-transparency-minecraft-26.3), the bench isolates the shape 2026-09-29 05:29:00 -04:00
swung0x48 47bbf5e6fd [Docs] (P10 A): CONTRACT-P10 section 1 - the fence-report protocol, the present-time report, the gates and red-once - with the re-scoped plan and A-FENCE-POLL, the paired host bench and the llvmpipe finding 2026-09-29 04:34:16 -04:00
swung0x48 73089cf2b3 [Test] (P10 B): the caps mirror's parallel-shader-compile entry is the client's switch in both directions - the extension-list read-back case pins compilation inline, and a new case holds the reconcile (red once with it removed) 2026-09-29 04:34:02 -04:00
swung0x48 80d205c401 [Refactor] (Remote, P10 B): the emit table's SetSwapInterval slot is the server forwarder - SetSwapInterval_Unmigrated and the BackendObject_Remote override deleted, class C empty (69 emitted / 0 unmigrated), and a unit control that the slot crosses as one control frame 2026-09-29 04:34:02 -04:00
swung0x48 4373964443 [Test] (P10 B): AsyncCompileScenario (AsyncOn/AsyncOff/OptimisticShaderStatus), XfbPrimitiveQueryScenario and PrimitivesGeneratedNoXfbScenario on the split arms of both backends - Espryt Split/Spawn/Tcp, Magma Split/Spawn/Tcp gated, PrimGenReroute split+spawn only 2026-09-29 04:34:02 -04:00
swung0x48 2271f6c1ec [Fix] (Remote, P10 B): GL_KHR_parallel_shader_compile follows the client's compile switch - the caps mirror re-derives it on adoption, so a tcp server started with another MOBILEGL_ASYNC_SHADER_COMPILE no longer decides the client's string 2026-09-29 04:34:02 -04:00
swung0x48 8d0a1b6ca5 [Test] (P10 A): PacedFrameFenceWaitsBench - the per-frame timed wait on an older frame's fence as an opt-in measurement (MGITEST_FENCE_PACE_BENCH, _LAG) on both backends 2026-09-29 04:29:49 -04:00
swung0x48 625f4db47d [Feat] (MG_Remote, P10 A): a present reports the fences finished so far before it returns the credit, so the client the credit releases finds the fence of a frame ago already known 2026-09-29 04:29:49 -04:00
swung0x48 b1d60523e4 [Fix] (Magma): an upload into an image the open recording already wrote, or that still has a queued clear, is submitted after that work - a glTexSubImage after a clear or a draw into the level no longer ends up under it 2026-09-29 04:28:55 -04:00
swung0x48 8966d5c294 [Fix] (Espryt, Magma): a partial write into a level whose upload is converted (widened RGB formats, Magma's RGB->RGBA expansion) or reshaped (Espryt 1D arrays) uploads only its dirty region - the rest of the stale shadow no longer lands over GPU-drawn texels 2026-09-29 04:28:22 -04:00
swung0x48 bcff9b00cc [Feat] (MG_Remote): fence polls are answered from the server's reports - the apply thread posts kEventFenceSignaled as the GPU finishes each client fence (and submits and re-checks while idle), glClientWaitSync(0) / glGetSynciv answer locally, and only every MOBILEGL_IPC_POLL_ESCALATE-th unanswered poll crosses 2026-09-29 03:59:44 -04:00
swung0x48 de33d40173 [Docs] (P9): close the phase - the contract's remint section and invariants, rulings ID-P9-11..12 with the integration gate, the roadmap row, the debts P9 hands on, the phase README and the status page 2026-09-29 03:36:45 -04:00
swung0x48 7aa00761f8 [Docs] (P9 W2): W2-REMINT - no pull protocol, the census, the coverage matrix and the red-once record; design/04 8.4, the P5c/P5e contract amendments, open question 2 and the P7 census attribution 2026-09-29 03:24:03 -04:00
swung0x48 aa0c242b51 [Test] (P9 W2): TextureRemintPullScenario - fourteen re-mint cases on both backends x monolith/Split/Spawn/Tcp, and the forced-readback-failure fallback entries 2026-09-29 03:23:54 -04:00
swung0x48 6c2f8a560c [Fix] (Espryt, P9 W2): a texture re-mint never needs the client - read back only levels the driver holds, replay the staged level when the readback is refused, count trp on the server path, and retire the texture-remint-pull markers and OnTexturePullRequest (callbacks 9 -> 8) 2026-09-29 03:23:54 -04:00
swung0x48 568090f09f [Fix] (Espryt): the image-bindable re-mint reads every level back from the driver texture before replaying it, so GPU-written levels survive the first glBindImageTexture - iris-derivative SSIM 0.996196552 -> 0.996376198 2026-09-29 03:06:45 -04:00
swung0x48 738b29f3e4 [Docs] (P9): the contract's first three sections (pack-buffer readback, graded log forwarding, the glGetError observation point), both fault injections, the W1 note with its red-once evidence, and rulings ID-P9-1..10 2026-09-29 03:03:03 -04:00
swung0x48 a7a82ffee5 [Test] (MG_Remote): P9 W3 - LogForwardTest leaves no role logs in /tmp on a green run 2026-09-29 02:42:20 -04:00
swung0x48 24560ffe81 [Fix] (MG_Remote): P9 W3 - a FATAL forward waits once per stall rather than once per line (MGLOG_F also marks survivable conditions), and F2 proves the flood got through on a live connection, inside TCP_USER_TIMEOUT 2026-09-29 02:42:20 -04:00
swung0x48 6bcfb1fba1 [Feat] (MG_Remote): P9 W3 - the server's log forward carries each line's level and never blocks a logging thread (<=WARN dropped and counted, >=ERROR lossless and paced, FATAL a bounded flush), fault injection F2, and the producerless OnLog callback deleted (9 -> 8) 2026-09-29 02:42:20 -04:00
swung0x48 362daafeed [Fix] (MG_Remote): the pack-buffer readbacks check their whole shape before the read and refuse it at one latch site per sink, each with a raw-peer control, and a verb counts as statically barriered only when every op that stamps it waits 2026-09-29 02:21:36 -04:00
swung0x48 6485ae1dc0 [Feat] (MG_Remote): a read into a pixel pack buffer crosses as read_pixels_to_buffer / get_texture_image_to_buffer and waits for nothing - the server lands the rows through the buffer-write gate on both backends, the client marks the buffer GPU-written, and a device lost with the write still owed ends in Fatal{ReadbackDeclined, "buffer-writeback"} 2026-09-29 02:02:29 -04:00
swung0x48 ad705a7392 [Docs] (P9): the implementation hand-off - the roadmap row checked against the tree (half of it is done, one design premise changed), five work packages and two fault injections 2026-09-29 01:23:58 -04:00
swung0x48 acfd367852 [Docs] (P12): the FCL fork's keep-running setting is committed and pushed 2026-09-29 01:10:19 -04:00
swung0x48 743f4e8e6e [Docs] (P12): close the phase - the contract, the disposition of every leftover item, a G1 line that only compares same-machine builds, and the doc citation the docs reshuffle broke 2026-09-29 01:05:45 -04:00
swung0x48 6e35ea2c02 [Docs] (P12): the portrait-window clipping was the 0x0 window request, not Magma - root cause, fix and evidence 2026-09-29 00:54:46 -04:00
swung0x48 c932af7f3c [Fix] (EGLImpl): a server-owned window surface with no EGL_WIDTH/HEIGHT asks for the native window's size, so a landscape game is no longer clipped in a portrait server window 2026-09-29 00:54:46 -04:00
swung0x48 8a71ea5954 [Docs] (P12): exit gate (a) passes - FCL's Minecraft renders into the render server window on both backends and killing the server latches device-lost in tens of milliseconds 2026-09-29 00:27:29 -04:00
swung0x48 984e2b69f4 [Feat] (tools): the FCL exit-gate driver - Minecraft renders into the render server window, then the server is killed 2026-09-29 00:27:28 -04:00
swung0x48 f2ad2fae9b [Fix] (MG_Remote): once the device-lost latch has spoken, declined verbs no longer log one ERROR line each - an FCL game wrote 113,080 of them in 48 s 2026-09-29 00:27:28 -04:00
swung0x48 27cbc53241 [Docs] (P12): exit gate (b) passes - a WSL client replays the in-world trace over Wi-Fi TCP at ssim 0.999883; the load frame is bandwidth-bound, the steady state is latency-bound 2026-09-28 23:22:06 -04:00
swung0x48 b60da6729d [Feat] (tools): the cross-host acceptance driver and its report reducer, and a note that netbench's nc sink understates the link 2026-09-28 23:22:05 -04:00
swung0x48 b4f0c53426 [Docs] (Disaggregated): the cross-host eglMakeCurrent hang was the host's xray TUN proxy stalling its relay, not MobileGL; direct Wi-Fi runs pass and are bandwidth-bound 2026-09-28 22:57:20 -04:00
swung0x48 429b4472bb [Fix] (tools): cross-host runs first prove the WSL -> phone TCP path is end to end, and netbench's pkill no longer kills its own shell 2026-09-28 22:57:20 -04:00
swung0x48 221a8144c2 [Docs] (Disaggregated): sync stage progress and roadmap with the five-arm perf comparison and the cross-host hang blocker 2026-09-28 22:33:53 -04:00
swung0x48 1987bdff5c [Feat] (Android): the render server screen and binary ship in the plugin APK too, not just trace 2026-09-28 22:10:56 -04:00
swung0x48 2fa58940e4 [Fix] (ci): the trace APK is a launchable app; assert exactly one launcher activity
The "Trace APK must not expose a launcher activity" check predates the render-server
screen: 1f72597dd gave the trace APK a launcher on purpose and 884f6aac9 moved the entry
icon back to PostActivity with a bottom navigation bar between the two screens, so the
apk.yml build job has been red ever since - taking the whole retrace matrix down with it
(skipped, not failed), which hid the distinction.

What is left of the old isolation contract is the count: exactly one
android.intent.action.MAIN in the merged manifest (a second one means a screen grew a
launcher entry by accident), and it belongs to PostActivity. The fclPlugin metadata ban is
unchanged.
2026-09-28 21:54:16 -04:00
swung0x48 35dddc0c89 [Fix] (MG_Remote): the socket-io stats are LinkMetrics free functions, not StreamLink members
6bccf98d7 put TakeSendStats/TakeReadStats on the concrete StreamLink class, so
ClientSession.cpp and ServerLoop.cpp had to include <MG_Remote/Transport/StreamLink.h> -
which the P6.5 seam gate (scripts/ci/link_seam_purity.py, "concrete link included above
Transport") rejects, and the MGPipe generators and hygiene gates job has been red since.

The counters stay in StreamLink.cpp beside the recv/sendmsg loops that feed them; only the
struct and the two Take functions move, to LinkMetrics.h as LinkIoStats +
LinkMetricsTakeReadStats/TakeSendStats. The readers drop the concrete include; the hot
loops are untouched.
2026-09-28 21:54:08 -04:00
swung0x48 3bdd565dc6 [Chore] (docs): Remove orphaned CS_Refactor notes accidentally committed in 4e48702f2 2026-09-28 21:16:35 -04:00
swung0x48 884f6aac96 [Feat] (Android): trace APK boots to POST; bottom nav bar switches to the render server
The trace flavor manifest removed PostActivity and made ServerControlActivity
the launcher, so the APK opened straight into the render server screen. Keep
both screens: PostActivity is the launcher again, and a bottom navigation bar
(NavBar, plugin flavor unaffected) switches between them via
FLAG_ACTIVITY_REORDER_TO_FRONT with transitions suppressed, so each screen
keeps its state. ServerControlActivity is singleTop now and handles extras in
onNewIntent so adb-driven starts still work. The POST screen also shows an
indeterminate progress bar while the self-test runs.
2026-09-28 21:06:44 -04:00
swung0x48 b9a89967b0 [Docs] (Disaggregated): five-arm transport perf comparison and bottleneck attribution
rd12/openra on-device matrix (monolith/inproc/spawn+shm/spawn+tcp),
tcp-localhost transport_wait decomposition (present credit=1 ~74% +
futex wakeup tax; SPIN_US=2000 alone 2.9x), cross-host control-plane
hang evidence chain blocking P12 acceptance B, Wi-Fi bandwidth ceiling,
and the prioritized knob/code optimization list.
2026-09-28 20:55:45 -04:00
swung0x48 141227e8ba [Feat] (tools): add the disaggregated-mode device perf matrix and profiling harness
run_matrix.sh drives interleaved monolith/inproc/spawn+shm/spawn+tcp
benchmark rounds on an adb device with per-run arm-marker verification;
profile_matrix.sh layers simpleperf, ftrace wakeup capture and a /proc
sampler on top; make_report.py aggregates runs into REPORT.md. The wifi
kit carries the cross-host control-plane hang repro and phone-side
strace/fd-reconcile helpers. .trace-work/ (run output) is now ignored.
2026-09-28 19:36:30 -04:00
swung0x48 930c29e9e0 [Feat] (Android): the render server screen takes intent extras for prefill and autostart
listen/token/env/onscreen prefill the form, --ez start true starts the
server, --ez stop true stops it - adb drives the screen without the IME,
which is how the on-screen unix path (@name + shm, ServerOwned window)
was verified end to end from the UI.
2026-09-27 08:36:57 -04:00
swung0x48 c16fb5a8ac [Feat] (MG_Remote): the in-process display server serves unix endpoints too
mobilegl_server_serve_inprocess refused anything but tcp:// because its
only supervisor was the TCP one's thread hand-off. A unix endpoint
(@abstract name or filesystem path) now runs a supervisor of the same
shape minus the TCP exposure machinery: AcceptPair pairs a client's
control and SCM_RIGHTS connections by arrival order, there is no
DataBind to route and no pre-auth gate, sessions run on threads one at
a time with Busy for a second client, the latch resets between sessions
and g_inProcessStop ends it. That gives the on-screen server the
on-device fast path: ServerOwned windows over a unix socket with shm
data, no TCP copies.

The control screen's tcp-only guard for the on-screen mode goes with
the refusal it mirrored.
2026-09-27 07:43:56 -04:00
swung0x48 e3e890ed5d [Feat] (Trace): the replay runner builds as an adb-shell executable for Android, and unix endpoints carry a headless window
MOBILEGL_BUILD_TRACE_REPLAY now also produces mobilegl_trace_replay on
ANDROID: the same CLI as desktop, linked against the APK runner's
glws/glproc glue. On the device it dials a render server with no
Activity at all - MOBILEGL_IPC_CONTROL=unix:@name with shm data to the
offscreen supervisor (a cross-uid abstract connect needs root or an
SELinux waiver; loopback TCP works from plain adb shell), or
tcp:// with MOBILEGL_IPC_SURFACE=server to the on-screen display server.

glws_android grows the desktop glws_egl's server-owned null-window
createWindowSurface, gated on MOBILEGL_TRACE_SURFACE=window so the APK's
pbuffer mode keeps its old answer, and the config carries EGL_WINDOW_BIT
for it. The control screen refuses unix endpoints in on-screen mode:
the in-process display server serves tcp:// only.

Verified on-device with an OpenRA retrace, ssim=1.0 both ways:
offscreen service on @mgl-cli with the CLI as root, and the display
Activity on tcp://127.0.0.1:40613 with the replay's frames on the
phone's screen.
2026-09-27 06:28:00 -04:00
swung0x48 bf6feebb22 [Perf] (MG_Remote): The producer's transport park is timed as its own per-frame number, and a reducer pairs the two role logs into a frame-time breakdown
transport_wait_ns measures wall time in Ring::Wait's flush/spin/sleep, so
the client's parked time is read directly instead of inferred by
subtracting CPU clocks that run concurrently with the server.
frame_stats.py matches client and server present serials into per-frame
avg/p50/p99 splits; the README documents the MOBILEGL_PIPE_STATS=1
workflow.
2026-09-27 04:45:07 -04:00
swung0x48 f10db410e8 [Feat] (Android): the render server screen also starts the on-screen display server
An On-screen window checkbox switches Start from the offscreen service
to MobileGLDisplayActivity (process :mglwin), which serves in its own
process and renders a client's window surface onto its SurfaceView.
Stop ends either shape: stopService for the supervisor, killProcess for
:mglwin, and the running check covers both. Verified end to end with an
OpenRA retrace: the client dialed tcp://127.0.0.1:40613 with
MOBILEGL_IPC_SURFACE=server from a freeform window while the display
Activity held the screen, frames landed on the SurfaceView and the run
passed with ssim=1.0.
2026-09-27 04:39:01 -04:00
swung0x48 aaa50c88e1 [Feat] (Android): the render server screen takes unix endpoints, and is titled for what it serves
The endpoint field accepts @abstract names and filesystem paths beside
tcp://, matching what mobilegl_server_main's Listen already binds - on
Android the abstract namespace is the on-device rendezvous that always
works, and a relative path lands in the app's files dir, the service's
working directory. Unix endpoints skip the token rule (same-device
reach by construction) and are probed with LocalSocket instead of a
TCP connect. The screen is now titled MobileGL Render Server.
2026-09-27 04:39:00 -04:00
swung0x48 1f72597dd8 [Feat] (Android): a launcher screen starts and stops the offscreen TCP server
ServerControlActivity (trace flavor) collects the listen endpoint, auth
token and extra env, then drives MobileGLServerService with the same
extras adb used. Non-loopback listens are held to the 16-byte token the
server enforces, and a connect probe confirms the supervisor bound its
socket. The screen is the APK's launcher icon, which the trace flavor
lacked since PostActivity is removed there.
2026-09-27 04:39:00 -04:00
BZLZHH 6bccf98d7f [Perf] (MG_Remote): Frame time is accounted on both sides, down to the socket reads and the client's waits. 2026-09-26 17:57:48 +08:00
BZLZHH 3a6c5c7f04 [Perf] (MG_Remote): The create window ships at 10, and the server now reports where its own frame time goes. 2026-09-26 15:30:48 +08:00
BZLZHH b017ec1c1f [Test] (MG_IntegrationTest/Scenarios): The P4a white-box applier reads fence on a readback. 2026-09-26 12:53:10 +08:00
BZLZHH 102f2ca7d5 [Perf] (MG_Remote/Client): The create window's shipped depth is 6, and its ceiling 8. 2026-09-26 11:26:54 +08:00
BZLZHH 1f1ece8a33 [Test] (MG_IntegrationTest/Scenarios): Fence the object-death gates on a readback instead of a clear. 2026-09-26 09:36:37 +08:00
BZLZHH 677689454a [Test] (P12): the window's gates move onto the arm it ships on - a real stream session with an answering peer, and it goes red by name when the store is a bounded ring
THE STRUCTURAL FIX, not just one more case. Three rounds of this feature were green in
the lanes while being inert on the device, because every window-level gate in the tree
runs in RemoteClientControls, whose session is TransportMode::InProcess - ShmLink - and
since P12 the window is OFF there by design. So the lanes and the device ran different
transports, and nothing in the tree could tell.

StreamClientPublicationTest.EveryDeferredCreateAnswerComesBackOnTheShippingArm runs on a
real StreamLink with a real ClientSession and a peer that ANSWERS: a scripted servo pops
each record, replies with its seq, advances and flushes the watermark - i.e. exactly what
the drain's WaitForApplied oracle needs, and nothing more (nothing applies a record).

IT IS PROVED TO GO RED, and by the right name. Reverting the store to "keep the last 8
arrivals" makes it die on the diagnostic added with the store:

    MGPipe: Fatal{ReplyMissing, "create window"} - seq 1 was declared and APPLIED, and
    its answer is no longer in the link's store; the window cannot advance past it and
    nothing recovers a lost answer

AND THE FIRST VERSION OF THIS GATE WAS WORTHLESS, which is the part worth keeping: with
24 creates and nothing else it passed against the ring as well, because between a push
and its read only a couple of answers arrive. What killed the device was TRAFFIC - the
server is behind and its late answers to earlier records fill a bounded buffer first - so
the case now emits twelve MORE records whose answers are wanted (declared
SetTextureParams rows) between the first create and the rest, waits for the peer to
answer them, and only then lets the drain come back for the create. Two things were
needed to make that deterministic: the settlement wait (otherwise it is a race the ring
can win by luck) and glFlush(), because a published record reaches the peer on a flush
and not on PublishAndNotify - the neighbouring case in the same file is built on that.

ALSO IN THIS COMMIT: the assertion is the CONSERVATION LAW, not "taken == creates".
Nothing drains the window when no further create arrives, so the last few answers stay
held - taken + pending == creates, with pending <= window, is the honest contract, and
taken == creates would be asserting a drain on a path the row does not have.

318 tests green across five lanes (172 + 15 + 22 + 103 + 6).
2026-09-26 03:41:48 +08:00
BZLZHH 3a2555805a [Perf] (P12): the create window lands as the default - declared-answer retention replaces the ring, and the load frame goes 28.66 s -> 16.52 s (11.02 s at window 4)
The three things that made the deferral inert are all fixed, in the order they were
found: the reply ring could not hold an answer across the ~1,000 records between two
creates; a residency budget could not help because what displaces an answer is what
the SERVER sends, not what the client emits; and so the mechanism had to be retention.

WHAT LANDED
  * ILink::DeclareReplyRead + LinkCapabilities::RetainsReplies. The wire declares a
    seq will be read BEFORE it publishes, so the declaration cannot race its own
    answer; a link with the capability false ignores it, and the window does not
    defer on one (CreateWindowEffective() is gated on it).
  * StreamLink's ring of the last 8 answers becomes a declared-answer store: a
    monotone whitelist plus a cursor, answers kept until their reader TAKES them
    (take-and-delete under one lock, bytes swapped into the scratch), and a capacity
    of 24 that is FATAL to exceed rather than evicting. An evicted wanted answer is
    unrecoverable - re-latching by sending another create wipes the server's record
    (PipeApply.cpp:2033) and its PendingUploads (PipeApply.h:332) while the client
    cleared its dirty flags at emission (PipeApply.h:298-304) - so the capacity is
    sized so it cannot be reached and reaching it says its own name.
  * An answer nobody declared is DROPPED, which is what keeps the store bounded by the
    client's deferral depth instead of by the server's answer rate. Retain-on-arrival
    grows ~11.3 entries per create against a server that ignores the no-reply bit.
  * A read that fails after WaitForApplied has returned Reached is now Fatal{ReplyMissing}
    naming the seq: on the stream arm the reply frame strictly precedes the progress
    frame that publishes appliedSeq, so that combination is a broken invariant, and
    retrying it (what the row used to do) wedges the window for the rest of the
    process while leaving every deferred object latched on a provisional accept.
  * The residency budget is deleted (WireTables.cpp keeps its history), the window
    retirement branch drains once before falling back - it could otherwise strand the
    answers it was holding - and Detach clears the store.

MEASURED ON THE DEVICE, load frame, both ends rebuilt, frame-sent identical per op
(create 4,536, records 103,097) throughout:

    window 1 (R-5's shape)              4,536 takes   28.66 s
    window 2 explicit                   2,500         17.27 s
    window 2 AS THE SHIPPED DEFAULT     2,563         16.52 s
    window 4                            1,644         11.02 s
    window 2 against an OLD SERVER      2,663         16.38 s   (zero Fatals)

The last row is the version-skew case the design review named as a must: a server
built before the no-reply bit answers ~55,903 records a frame, the whitelist drops
every answer nobody declared, and the deferral still reads its own back.

GATES (172 + 15 green), and the important one is proved to go red:
  * ADeclaredAnswerSurvivesFarMoreInterveningAnswersThanAnyFixedBuffer - reverting the
    store to "keep the last 8 arrivals" fails it immediately with "the declared answer
    did not survive 64 intervening ones". That is the line between the ring-1024
    stand-in and a bounded mechanism.
  * AnAnswerNobodyDeclaredIsNotKeptAtAll - red if everything is retained.
  * TheWindowRefusesToDeferOnALinkThatCannotRetainAnswers - on the in-process (ShmLink)
    arm the knob at 4 must still block every create, red if the capability gate goes.
  * The refused-create case now reads the repair that exists on THAT arm (the latch),
    and asserts the window's own books are empty there.

NOT DONE, and named: the window has no unit gate on the STREAM arm yet - only the
device runs. Three rounds of this feature were green in the lanes while inert on the
device because the lanes and the device run different transports; the transport-level
retention gates close that for the store, and the window-level one should be built on
StreamClientPublicationTest's real stream session next.
2026-09-26 03:31:14 +08:00
BZLZHH 1ef7439625 [Perf] (P12): with the budget off and a deep reply buffer the window pays - 28.66 s -> 11.45 s at W=4, and the round trip does NOT grow, so the ~5 ms is a wakeup the server batches away
The second one-line experiment, and the one that decides whether the retention design
is worth building. Ring 1024 plus the residency budget DISABLED (its constant raised
out of reach), W=2 and W=4:

  arm                        waits   rtt_mean   by_op 2    wall
  W=1 (today's default)      4,543   5,672 us     4,536   28.66 s
  budget on + ring 8         9,078   2,598 us     9,071   27.04 s
  ring 1024 + budget 7       3,450   5,367 us     3,443   23.04 s
  ring 1024 + budget OFF W=2 2,560   4,916 us     2,553   16.80 s
  ring 1024 + budget OFF W=4 1,691   4,224 us     1,684   11.45 s

rtt_mean does NOT grow (4,916 / 4,224 us, both BELOW the 5,672 us baseline) while the
waits fall by 2x and 2.7x and the wall follows. So the ~5 ms is a WAKEUP round trip and
the server drains a batch per wakeup - which is exactly what makes W pay, and it is the
question the audit listed as undecidable from the repo and the watershed for whether
this is worth doing at all. It is.

AND IT BEATS THE HANDOFF'S OWN FLOOR: handoff-2 section 0.4 put the window's landing at
15-20 s, treating the client's ~13.4 s CPU as ADDITIVE with the waits. This frame turns
out to be wait-dominated, so amortising four creates per wakeup lands at 11.45 s, 3.5 s
below that estimate.

1024 IS A STAND-IN, NOT THE DESIGN: it works only because 1024 happens to be deep enough
for this workload, and 1024 retained answers is ~2 GiB worst case (a ReadPixels answer can
be 2 MiB). Retention gives the same property - the answer is still there - with a bound
tied to the client's own deferral depth, which is why the implementation the audit
specified is what gets built, not "make the ring 1024".

Also corrects the note's section 7 path 1 (audit F15): on the stream arm SEG_REPLY and the
Hello's ReplySlotCount have nothing to do with answers - they travel as Reply frames and
StreamLink::ReadReply reads the client's own store - so "deepen the pool" does not address
the arm that was measured, and should not be paid for.

Reverted (ring 8, budget 7, probes deleted); 172 + 13 green.
2026-09-26 03:01:20 +08:00
BZLZHH 82de3f6826 [Docs] (P12): the ring-1024 experiment - every deferred answer comes back (readfail 8,872 -> 0), so eviction was the whole cause and the residency budget is now the brake
One-line experiment before writing the retention table: StreamLink's kReplyRing 8 -> 1024,
nothing else changed, same load frame on the same device.

                       ring = 8 (today)   ring = 1024
    readfail                    8,872              0
    taken                          63          4,499 / 4,500
    blocked (fell back)         4,435              0
    by_op 2                     9,071          3,443
    wall                        27.9 s         23.04 s
    frame-sent        create 4,536, records 103,097   identical per op

TWO CONCLUSIONS. First, eviction was the ONLY cause: with a buffer deep enough, every
deferred answer was read back and not one create fell through to the blocking path.
The wedge (want=310) and the useless budget (budget=4,451) were two faces of "eight
slots is not enough". Second, what brakes now is the BUDGET, not the buffer: by_op 2 =
3,443 matches the number of times the budget forced an early read (3,395), i.e. every
one of those is a real blocking wait. Its constant is 7, sized for the 8-slot pool, so
against a 1,024-deep buffer it became the only throttle and the wall only moved 27.9 ->
23.0 s.

WHAT IT MEANS FOR THE DESIGN. Once the answers are RETAINED until read, correctness no
longer depends on how long one survives in a bounded buffer, so the budget must be
sized to the RETENTION CAPACITY rather than to the pool - or be replaced by "retained
means safe". by_op 2 = records/W then puts W=2 at ~2,268 waits and the wall at ~15 s,
which is the goal's own success criterion.

Experiment reverted (ring back to 8, probes deleted); 172 + 13 + 22 + 103 green.
2026-09-26 02:44:30 +08:00
BZLZHH b1ef484116 [Perf] (MG_Remote): the residency budget does not make the window usable either - the answer is gone before the client next looks, so a deferral needs RETENTION, not a budget
Steps 1 and 2 of the plan the design review set out, landed and measured on the
device. Step 1 answers the question step 2 depended on; step 2 answers the one the
whole line depended on, and the answer is that the window's shape is wrong.

STEP 1 - CAN AN ANSWER BE LOST IN DELIVERY? NO. ReplyPool now counts what the link
took (PostedReplies) and what it refused (FailedReplies), and a non-OK result is
NAMED instead of swallowed: it checked only BUFFER_TOO_SMALL and let
TRANSPORT_CLOSED vanish with no line and no counter, so "the server never sent it"
and "it was sent and the buffer overwrote it" were the same symptom with opposite
fixes. Measured: posted=4501 failed=0. The twenty-answer gap between the server's
tally and the client's was a sampling difference, not a drop - and the server's
m_answered counts ATTEMPTS (incremented before PostReply), which is why it could
never have settled it.

STEP 2 - DOES A RESIDENCY BUDGET FIX THE WINDOW? NO.
Wire_ResourceCreate now measures how many reply-owning records the client has
emitted since the oldest window entry was pushed (ClientSession::ReplyPostings, a
new clock) and takes the oldest answer back as soon as that approaches the reply
buffer's depth, instead of deferring past the point where the answer can still
exist. On the device:

    P12win call=4501 cfg=2 eff=2 pending=2 taken=63 pushed=65 blocked=4435
           budget=4451 readfail=8872 posts=54377
    kind=frame-op frame=14 wait_replies=9078 by_op=2=9071   wall 27.87 s

budget=4451: it fires on essentially EVERY create, the window stops deferring
(pushed=65), and the drain still reads nothing back. THE REASON IS STRUCTURAL, not
a constant:

  * the window only gets control WHEN A CREATE ARRIVES. Between two creates this
    client emits on the order of a thousand records - at the FIRST failed read the
    reply ring already spanned [1175..1504] while the head was seq 310 - and nothing
    looks at the window in that gap;
  * what displaces a deferred answer is not what this client emits, it is WHAT THE
    SERVER ANSWERS. The server is the one behind (one answer per ~5 ms wakeup while
    the client runs ahead), so the replies arriving during the gap are its LATE
    ANSWERS TO THE EARLIER CREATES - they fill the 8 slots first;
  * so the check fires too late by construction. Tightening it does not fix that; it
    only removes the deferral entirely, which is what pushed=65 is.

WHAT THAT LEAVES. A deferral is readable only if the client KEEPS the answers it
deferred. The buffer is bounded by COUNT and the client cannot bound how many
answers the server will post between two of its creates, so hoping the buffer still
holds one is a race the client loses - which is also why the earlier two diagnoses
(reply-pool churn, then FIFO-with-no-way-out) were both describing symptoms of the
same thing. The shape that can work is retention: the reader keeps the answers for
the seqs still outstanding (a handful), drops them when read, and reports a LOSS
when a read asks for one it no longer has. That is a transport change - StreamLink's
ring becomes a retention map with a cap and a lost counter - and it is the next step.

BOTH STEPS STAY: step 1 is a net gain on its own, and step 2 is the safe direction
(it can only degrade the window to no-deferral, i.e. the pre-window shape) and its
clock is the one the retention design needs. MOBILEGL_IPC_CREATE_WINDOW stays at 1.

172/172 and 13/13 after the probes were removed. The measurements and the revised
conclusion are in notes/p12/CREATE-WINDOW-MEASURED.md section 9.
2026-09-26 02:35:05 +08:00
BZLZHH 4e48702f26 [Feat] (MG_Remote): a record can tell the server its answer will never be read - the bit works and cuts a frame's replies by 92%, and it proves the window's real blocker is the FIFO drain, not the pool depth
The create window could not collect its deferred answers. The previous commit
diagnosed that as the reply pool being 8 slots deep against 55,903 replies in one
frame, and left a no-reply bit as the way out. This lands that bit, measures it on
the device, and reports that the diagnosis was half wrong.

WHAT LANDED. ClientSession::EmitAndWaitTails takes a second question beside
wantReply: willReadReply. They are not the same question and the window is the case
that separates them - it emits with wantReply=false and READS the answer later, so
its records must still be answered. kRecNoReply (RingRecordFlags bit 5, which
aliases kOptional in the call-flag space; the flag-space asserts were extended and
the alias named) is stamped when willReadReply is false, and the decoder skips
PostReply while keeping every acceptance counter - "nobody reads the answer" is not
"nothing happened". The two illegal combinations are Fatal by name.

Three call sites say it: SetTextureParams for a confirmed object, the respecify
row's fire-and-forget half, and sub-data where MGPipeSubDataWantsItsReply decides.

MEASURED ON THE DEVICE (aikjs4s8cys8ba7l, rebuilt APK + client, both halves new):
    server:  skipped=50,001   answered=4,521        (a frame's replies: 55,903 -> 4,543)
    client:  Reply frames received seen=4,501, latest seq=97,181
    frame-sent: identical per op, records untouched
    create row: by_op 2 = 9,070 (was 4,536)  - because the drain's blocking take is
               now counted, so the row's number stays honest instead of reading as
               "create stopped waiting" while the waiting merely moved
AND THE WINDOW STILL READ BACK NOTHING: readfail 8,951 -> 8,872, i.e. -0.9%.

THE REAL CAUSE, from the miss probe:
    P12miss n=1     want=310 replySeq=1504   ring=[1175..1504]
    P12miss n=9001  want=310 replySeq=102915 ring=[102866..102915]
want=310 every time. The drain is STRICTLY FIFO and a failed read has no way out -
it retries the same head for ever. Any reply buffer is bounded by design (8 pool
slots, or the 8-frame ring the stream link now keeps), so with 4,536 deferred
answers at least one is overwritten before it is read, and that one wedges the
window permanently. Cutting the churn by two orders of magnitude does not help: it
turns a certainty into a certainty. That is the constraint, not the depth.

THE FIX IS NAMED AND NOT TAKEN (the goal this work ran under stops here on purpose).
On a failed read of the head, abandon it: MGPipeNoteHandleUnpublished +
NoteCreateSuspect + pop - the DECLINE path reused for "unknown". The window then
self-heals through the blocking path and the object merely gets slower. It needs its
own gate: a failed read must let the window ADVANCE.

Gates, all green (172 + 13 + 22 + 103):
  PipeWireCodecTest.TheNoReplyBitSkipsTheAnswerAndKeepsTheAcceptanceRecord - the skip,
    the kept tally, the answered/skipped split, and the negative control that a record
    WITHOUT the bit is still answered (a server that stopped answering everything
    would otherwise pass)
  RemoteClientControls.ACreateWindowDefersTheAnswersAndTheWindowOffDoesNot - now also
    asserts the answers CAME BACK (answers_taken=11, outstanding=1, fallbacks=0 against
    12 records). RED-ONCE PROVED: giving the window's creates the literal
    "fire-and-forget => skip" predicate makes this case fail.
  RingTest - the sixth and last flag alias, and the hand-written exhaustiveness list
    extended (with a note that it is only as exhaustive as that list).
  Server side: AnsweredRecords()/SkippedAnswers() - the number that made this
    debuggable, since a wire change whose effect is invisible is undebuggable.

Client-first is safe: an old server ignores bit 5 and replies as before, so the two
cross-version directions degrade to today's behaviour - this is "both ends must
upgrade to take EFFECT", not "to avoid breaking". Hit that in practice first.
2026-09-26 02:00:44 +08:00
BZLZHH 7c822cdc85 [Docs] (P12): the no-reply flag reviewed - the arithmetic holds and the flag is cheap, but the predicate has to be "this answer will never be read", not "fire-and-forget"
An independent review of the one design door the measurement left open (stop sending
the ~51,000 answers nobody reads, so the 8-slot pool stops churning and the create
window becomes collectable). Verdict: feasible, worth doing, and the proposal AS
WRITTEN would disable the thing it exists to enable.

THE CATCH. The windowed create is wantReply=false AND its answer is read -
WireTables.cpp's drain reads exactly those seqs. A bit keyed on "fire-and-forget"
therefore stops the server answering the creates the window is waiting for: the
window goes from "cannot be collected" to "there is nothing to collect", and the
readfail counter that is the only way to diagnose it disappears with it. It fails
silently, not loudly. The predicate has to be per-record "this answer will never be
read" - true of params / respecify-escape / sub-data, false of a windowed create -
carried as a separate input from wantReply (willReadReply), with
willReadReply && !rowCarriesReplySlot rejected rather than ignored.

VERIFIED HERE, not taken on trust (the reviewer's file:line claims, checked):
  - the flag space is RingRecordFlags (Ring.h:281-285), bits 0-4 taken, BIT 5 FREE,
    and the header's own table (PipeWire.inc:35-65) says "bit 5 kOptional vs (nothing)
    no ring counterpart" - so a new ring bit aliases kOptional, which is exactly what
    the assert at PipeWireCodec.cpp:142-145 refuses;
  - the three static_asserts that will trip are real and deliberate
    (PipeWireCodec.cpp:106, :142-145, RingTest.cpp:740);
  - WireFingerprint mixes payload-layout, catalogue, render-state digests + OpCount,
    none of which a ring bit moves - so no protocol revision bump is required, and
    both cross-version directions degrade to today's behaviour. That makes this
    "must upgrade both ends to take EFFECT", not "must upgrade both ends not to
    BREAK" - one notch cheaper than handoff-2 section 3.2's protocol surface;
  - StreamLink.cpp:564-571 is the independent progress flush (>= 64 records or
    >= 1 ms) and :577 is PostReply piggybacking SendProgress on the reply - so
    skipping replies coarsens progress granularity. Bounded, but unmeasured.

AND IT CORRECTED TWO OF MY OWN NUMBERS. The frame's reply count is 55,903, not
59,673: op 4 ResourceDestroy carries kNone in the catalogue (PipeWire.inc:179) and
PostReply Fatals on a row with no kReplySlot, so it never answers - I had added its
3,770 records to the total. The derived figures move with it (12.3 replies per create,
0.65 of a create of survival, 6,988 pool churns), and every copy is fixed here:
Config.h, ConfigLoader.cpp, WireTables.cpp's section comment, CONTRACT-P5E.md,
CREATE-WINDOW-MEASURED.md.

TWO MORE THINGS THE FOLLOW-UP MUST CARRY, both recorded in the note's section 7:
  - LinkMetricsBeginReply counts ownsReplySlot, so once the flag lands by_op 2 falls
    to ~0 - and section 1 of the note tells the reader to use by_op 2 to check whether
    the window engaged. That check becomes a FALSE GREEN; the flag needs a
    server-side "how many times did this op answer" counter beside it.
  - the ~112,000 sendmsg/frame the flag removes are on the apply thread, so the
    flag's own value is a measurable claim that nobody has measured.

172/172 and 13/13 after the comment corrections.
2026-09-26 01:13:31 +08:00
BZLZHH 7205509a9a [Docs] (P12): the create window is dead by arithmetic, and the two debts it exposed (the applier can no longer say it dropped a record; the picture still has no real control) 2026-09-26 00:51:56 +08:00
BZLZHH d011395939 [Docs] (MG_Test, P12): the dropped-records repair is gone, not merely narrowed - the first-use work removed the last row that read an applier refusal, and the case that pins the debt
handoff-2 section 3.3 asks for a gate on the one scope that empties the applier's
object records while the frontend objects live on (MGPipeApplierReleaseObjectRecords,
"latent" because its only caller in the tree is MG_Test). The gate was expected green
for a CHANGED descriptor: the respecify is refused, RespecifyOnce turns the refusal
into a create plus one retry. It is not green.

MEASURED:
    dropped-records probe: creates=0 applier_records=0 published=1

Nothing is emitted at all. The refusal can only reach the emitter through an answer
the row READS, and P12 made both rows answer themselves for a "confirmed" object:
the respecify is fire-and-forget (MGPipeResourceRespecifyWantsItsReply) and the
params row writes a PROVISIONAL kStatusOk (Wire_SetTextureParams). "Confirmed" is
MGPipeHandleIsPublished, which answers "did a create GO OUT" - and it stays true
after the records are dropped, which is the scope's whole definition. The only row
left that reads the real answer is create, and create is only sent when the latch is
clear - the one thing that stops being true.

So section 3.1's "clearing the latch is not enough" is the smaller half: nothing
clears the latch either, and nothing can until the applier can say "I hold nothing for
this handle" - section 3.2's third option, the reverse channel, i.e. the protocol
surface the handoff hoped to avoid paying for. That hope was priced on the trigger
being unreachable; it is still unreachable (no production caller), so the call stands,
but the cost of a production caller is now known to be the whole thing.

THE CASE IS WRITTEN AS A DEBT RATHER THAN A SPEC, and says so in its name and its
comment: it pins today's three numbers, so whoever lands the signal sees THIS case go
red and knows the repair started working. Removing the heal from RespecifyOnce would
not make it red - nothing reaches the heal - which is exactly why the assertion is on
the applier's table and the latch rather than on a count (handoff-2 section 3.3: "a
count cannot show this class of error").

Also rewritten: notes/p12/UNPUBLISH-GUARD-SIGNALS.md sections 4 and 6 now carry this
as 4.2 (sections 2 and 3 stand: neither caps generation nor FreshlyPrimed is a signal).
172/172.
2026-09-26 00:50:00 +08:00
BZLZHH d2efd1d79d [Perf] (MG_Remote/Client): the create window lands OFF - the deferral cannot be collected, 8 reply slots against 59,673 replies in one frame - measured on the device at 4,536 blocking creates either way
The window was the handoff's remaining item: let up to N resource_create answers be
outstanding, read a provisional accept, collect the answers in order, and un-latch an
object whose late answer is DECLINED. It is a sound mechanism and the gates below pin
it - and on the device it does nothing at all, which is why the default is 1.

MEASURED, same device, same frame, same 101,421,396 B staged, same binary:

    MOBILEGL_IPC_CREATE_WINDOW=4   by_op 2 = 4,536   wall 28.64 s
    MOBILEGL_IPC_CREATE_WINDOW=1   by_op 2 = 4,536   wall 28.64 s
    (the handoff's pre-window baseline: 31.3 s)

by_op 2 counts the creates that took the BLOCKING path (LinkMetricsBeginReply is
called with ownsReplySlot = rowCarriesReplySlot && wantReply), so it not moving means
no create deferred. A temporary counter in the row said why: pushed=24, blocked=4,476,
took=20, readfail=8,951, pending pinned at 4 - the drain READ the answers back and
failed every time.

WHY, IN THE PROTOCOL'S OWN ARITHMETIC. An answer is readable only until its pool slot
is reused, and the pool is kDefaultReplySlotCount = 8. The server answers EVERY
reply-owning record, not only the rows that wait, because it cannot know which those
are (ClientSession.cpp:1897-1900: "The server posts either way"). This frame posts
59,673 of them: create 4,536 + respecify 20,633 + params 18,056 + sub-data 12,671 +
the rest. 8 slots against 59,673 replies churns 7,459 times per frame, i.e. 13.2
replies between two creates, so an answer is overwritten after 8 replies - 0.61 of a
create. The deferral is not slow, it is UNREADABLE, and no window depth fixes that.

AND IT IS NOT NEUTRAL: with the window on, the provisional accept latches an object
the applier may have refused, and the un-latch that corrects it lives in that same
unreadable drain. A refusal would become an object the client believes is published -
the silent divergence the suspect list exists to prevent. So the machinery is kept
(one experiment away from usable: a deeper pool, or a "do not answer me" bit, both
wire changes) and the default is off.

Worktree note: the window implementation itself was uncommitted in the tree at
82ed1df9 (MOBILEGL-CS-HANDOFF-2.md section 2, "can compile, gates mostly green, never
run on a device"). This commit lands it in the state the measurement supports. The
row also gains the heal the suspect list was missing - a BLOCKING create that lands
takes its object back out of it, or kCreateSuspectMax recoveries would retire the
window for the rest of the process - and the clamp that ties the environment's value
to the array, with a static_assert tying the array to half the reply pool so "someone
raises kCreateWindowMax to 8 and steps on the pool" is a compile error.

Gates (RemoteClientControls.inc), all green, 171/171 RemoteClientTest:
  ACreateWindowDefersTheAnswersAndTheWindowOffDoesNot - window on = 12 records 0 waits;
    the SAME CHILD turns the knob to 1 and it is 12 records 12 waits (the negative
    control); pending <= effective after every create; 99 clamps to the array; the
    array is <= half the pool
  ARefusedCreateUnlatchesItsObjectAndItsNextCreateBlocks - a real refusal (the
    server's consumed mask narrowed) clears the latch, +1 suspect, the object's next
    create BLOCKS (2 records 2 waits), and restoring the consumer HEALS it (latch
    back, suspect -1)

Docs: CONTRACT-P5E.md section 2.5 rewritten (it still said all three rows keep
kWaitReply), plus
docs/Disaggregated/notes/p12/CREATE-WINDOW-MEASURED.md (the numbers, the arithmetic,
the picture evidence: both arms SSIM 0.977) and
docs/Disaggregated/notes/p12/UNPUBLISH-GUARD-SIGNALS.md (handoff section 3: neither
caps generation nor FreshlyPrimed is a signal for "the applier dropped its records",
and the residual gap is one line - the dedup at TextureEmit.h:838 does not read the
latch, so an UNCHANGED descriptor after a drop produces no record and therefore no
refusal to self-heal from).
2026-09-26 00:47:19 +08:00
BZLZHH a2c1f5f292 [Fix] (MG_Remote/Transport): StreamLink::ReadReply answers by seq instead of "is it the newest" - the contract ShmLink already kept, and the reason a deferred answer could never be read back
ILink::ReadReply takes a SEQ (ILink.h:285) and ShmLink implements it against a
ReplySlotPool, so any answer the pool still holds is readable. StreamLink kept ONE
reply - replySeq/replyStatus/reply - and then asked the wrong question of it:

    x.replyReady.wait(lock, [&]{ return closed || x.replySeq >= seq || appliedSeq >= seq; });
    if (x.replySeq != seq) return ... MOBILEGL_ERR_PROTOCOL_MISMATCH;

The wait is written as "at least this seq"; the check is "exactly this seq". Every
answer older than the newest was therefore unreadable, which nothing noticed because
every caller read the answer it had just waited for - i.e. always the newest one. It
is the transport half of the P12 create window's failure: the window defers up to
four answers and collects them in order, and this link could not hand one back.

THE RING, AT THE POOL'S OWN DEPTH. kReplyRing = kDefaultReplySlotCount (8), because
that is the pool this link stands in for: an answer a reader has not taken by the time
eight newer ones have arrived is one the shm arm has already overwritten too, so the
two transports now lose the same answers and no more. Vectors are sized per reply, so
an idle link pays eight empty ones.

This does NOT make the create window work - on the device it changed nothing, because
the pool itself is 8 slots against 59,673 replies in one frame (see the next commit).
It is kept because the two transports disagreeing about ILink's contract is a bug in
its own right, and because the single slot silently failed any read that raced a newer
reply.

Gate: StreamLinkTest.AnOlderReplyIsStillReadableByItsSeq posts three replies and reads
them oldest-first - the order the window's drain takes them in.
2026-09-26 00:47:06 +08:00
BZLZHH 82ed1df90e [Perf] (MG_Pipe, MG_Remote/Client): a texture respecify for a CONFIRMED object stops waiting - the row pays its round trip once, at first use - and the frame drops from 130.9 s to 31.3 s
- The predicate is one function again: MGPipeResourceRespecifyWantsItsReply (PipeRoute.h) is read
  by the wire emitter, so the two halves of this row cannot disagree about whether an answer is
  owed. The buffer half never owes one (both call sites discard the Bool, PipeFill.cpp:1105/:1142);
  the texture half owes one exactly until MGPipeHandleIsPublished, and that latch is set only by an
  ACCEPTED create - which EmitResourceRespecify already sends, blocking, when it is clear
  (TextureEmit.h:869-879). So the object pays once, at first use, and every record after it is
  fire-and-forget.
- MEASURED on the device, same frame, same 101,421,396 B staged, stock power policy: waits
  25,154 -> 4,543 and wall 130.9 s -> 31.3 s. The op split is the proof of what changed and what
  did not: waits by_op 3=20611 -> absent (only 2=4536, 5=6, 9=1 remain), while the RECORDS sent
  are unchanged - frame-sent still shows 3=20633 and 103,094 records total. The round trips went
  away; the work did not.
- It also lands on an independent prediction: an audit run beside this work (mgl-device/scratch/
  scheme-review.md) computed 4,543 remaining waits and a 24-37 s credible range (midpoint ~30 s)
  from the archived logs, a host-inproc round trip of 35.4 us and a measured 43.1 MiB/s link.
  Measured: 31.3 s, at the midpoint.
- Cumulative on that frame since this work started: 43,232 waits / 221.5 s -> 4,543 / 31.3 s (7.1x).
  What is left is the create row (4,536 waits, ~23 s) - the confirmation itself, not waste - plus
  hard floors of ~13.4 s client CPU and ~2.2 s of transfer. Going below that means batching the
  create row, which keeps real answers.
- TWO DEBTS, STATED: (1) the invalidation guard for the one silent-loss case is NOT wired; the
  audit confirmed independently that no production path can drop a texture record today, and if
  one appears the notice it needs is protocol-surface (a tenth reverse callback). (2) No SSIM or
  visual check has been taken since these rows stopped asking - frame-sent agreement is strong
  evidence but it is not a picture.
- Two stale controls in older cases were fixed rather than left: the item-B gate now names the
  CREATE row (its old total-based control only moved because a respecify was still waiting), and
  the by-op measurement asserts records-sent for the rows that deliberately stopped waiting.
  RemoteClientTest 169/169; red-once for the predicate is forcing it back to "always wait".
2026-09-25 23:29:08 +08:00
BZLZHH f2b55b9d2d [Perf] (MG_Remote/Client): the buffer half of resource_respecify stops waiting - its answer is discarded at both call sites - and the measurement says the texture half is where the other 20,611 waits live
- The two halves of this row are not the same row. PipeFill.cpp:1105 and :1142 both call
  MGPipeRouteResourceRespecify(desc, nullptr) and drop the Bool, so for a buffer the wait bought a
  value that was thrown away. It goes fire-and-forget with the provisional-accept shape
  resource_subdata has had since item B.
- The texture half keeps its answer ON PURPOSE: there the Bool is the self-heal trigger AND the
  gate on the acceptance mirror, and TextureEmit.h:885-888 (ID-18 M3) says advancing that mirror on
  an unconfirmed answer is how the next identical call gets suppressed "against a record that was
  never stored". That half needs the provisional/really-accepted distinction plumbed through the
  reply mailbox before it can stop waiting; it is not a flag flip.
- MEASURED, same frame, same 101,421,396 B staged: waits 25,176 -> 25,154 and by_op 3=20633 ->
  3=20611. So the buffer half is worth 22 waits (0.1%) and the remaining 20,611 are texture
  respecifies - which is what says the next step is the mirror design, not another flag. Wall
  134.4 s -> 130.9 s.
- The ERROR-is-not-a-decline case moves to the texture half: a row that reads no answer cannot fold
  ERROR into a decision, so the rule it exists for belongs to the half that still waits (the
  transport fault is caught at the next row that does). New case
  ABufferRespecifyTakesNoReplyWhileATextureOneStillDoes pins both directions; red-once with the
  branch removed is exit 155. RemoteClientTest 169/169.
2026-09-25 22:53:38 +08:00
BZLZHH 5ad8370b37 [Docs] (P12): the first-use confirmation measured on the device - 43,232 waits to 25,176, 221.5 s to 134.4 s, op 47 gone and every other row identical 2026-09-25 22:36:51 +08:00
BZLZHH b62709604d [Perf] (MG_Impl/Pipe, MG_Remote/Client): a parameter change for an object whose existence is confirmed takes no reply - the round trip moves to the object's FIRST USE
- The device run put 18,056 of one atlas frame's 43,232 round trips on this row, at ~5 ms each, for
  records carrying a few hundred bytes. The acceptance it waited for says exactly one thing - "the
  applier holds a record for this handle" - and the client already keeps its own copy of that fact
  in MGPipeHandleIsPublished, which is set when the create was ACCEPTED and is never cleared while
  the record lives (the applier keeps texture records across a make-current, PipeFill.cpp:3756-3759).
  The wait was the client asking the server a question it had already answered.
- SO THE CONFIRMATION MOVES TO THE FIRST USE. EmitTextureParams now checks the latch BEFORE it
  sends: an unpublished object gets HealUnpublishedRecord - the same create-with-no-storage, the
  same storage respecify, the same one re-send the refusal branch has always run, extracted so both
  callers share it - and that create is the ONE blocking record the object owes. Every parameter
  record after it is emitted fire-and-forget and the caller reads a provisional accept, exactly as
  resource_subdata has done since item B. Wire_SetTextureParams reads the same latch to decide, so
  the two halves cannot disagree about whether a wait is owed.
- THE FALLBACK IS THE OLD PATH, unchanged: an object whose create was refused stays unpublished, so
  its next parameter record still takes the real answer and the existing one-retry heal still runs.
  That is also the red-once: with the unconditional EmitAndWait put back, the new case fails -
  RemoteClientControls.AParameterChangeForAConfirmedObjectTakesNoReply exits 153, having asserted
  that the eight parameter records still go OUT while none of them takes a reply.
- WHAT IS GIVEN UP, stated rather than implied: a loss the client believes is published (a bug, or a
  served-context teardown whose only caller today is a test) is now silent on the CLIENT - it stays
  counted and named on the server, which is the same residue item B accepted for the upload row.
- The measurement case's "the counter is live" control moves to create/respecify, because this row's
  zero is now the fix rather than a dead counter. 168/168 in RemoteClientTest; the unit lane is
  2617/2617. Tracks: docs/Disaggregated/notes/p12/DEVICELOST-AND-UPLOAD-WAITS.md.
2026-09-25 22:31:16 +08:00
BZLZHH c05e702d8c [Docs] (P12): the server's spin budget is the portable half of the idle-exit fix - 221.5 s -> 182.8 s on the same frame, no root and no code change
- 2.9 established that a record's ~5 ms is mostly the DEVICE waking out of a deep idle state.
  That points at a knob that is already wired: how long the server's apply thread spins before
  parking. Raised from 50 us to 2000 us (via the Activity's existing `env` extra, which
  ServerEnvironment.apply parses as KEY=VALUE;KEY=VALUE and which does not scrub
  MOBILEGL_IPC_SPIN_US), the same 43,232-record frame goes 221.5 s -> 182.8 s (-17%), with
  rtt_mean 5,040 -> 4,145 us. Pinning the device and disabling its deep idle states is worth
  166.5 s (-25%) but needs root; the spin needs neither root nor a rebuild.
- The server log proves the knob landed (`mgl-srv-apply started ... spin 2000 us`) rather than
  leaving the reader to trust an env var, and the note carries the exact am start line.
- So the load stall now has three ordered levers with numbers behind them: batch the waits
  (client-only, not built), keep the server hot (this commit, usable today, trades device CPU for
  wall clock on a frame that is 94% waiting), and pin the device for measurement discipline.
2026-09-25 21:36:39 +08:00
BZLZHH 7c72dfde2b [Docs] (P12): the ~5 ms per round trip is the DEVICE's deep-idle exit, not the link and not the client - which is what the handoff measured and could not explain
- Three one-variable experiments on the same device, same workload (the 43k-record frame): a 40x
  larger client spin moves the per-wait mean by ~10% (so it is not the client park/wake); the
  same frames over adb forward - a link whose raw RTT is several times lower - by ~25-30% (so it
  is not the link, which is what the handoff saw in 3.6 and could not account for); and pinning
  the device (governor performance on all eight cores, cpuoff_l / clusteroff_l / mcusysoff /
  s2idle disabled) by 26% per wait and 25% of the whole frame: 221.5 s -> 166.5 s for 43,218
  records and 100.5 MB staged, against 43,232 and 101.4 MB in the baseline.
- The mechanism: the server's apply thread parks between records, the SoC drops into a deep idle
  state, and every record pays the exit latency - milliseconds, once per record, which is the
  entire shape of the frame. It also matches the server profile the handoff took (65% socket
  syscalls, ~15% doorbell spin: waiting and waking, not working).
- Stated where it matters: the handoff's 249 s was taken on an UNPINNED device, so its absolute
  numbers carry this idle tax - the project has a pinning discipline for exactly this reason.
- The lever list is now ordered: batch the waits (K <= 8 by the reply-slot pool, client-only),
  keep the server hot between records (its own spin budget, which the APK does not wire yet), and
  measure split numbers on a pinned device. The device was restored to stock afterwards.
2026-09-25 21:28:13 +08:00
BZLZHH 6e7c0ce710 [Docs] (P12): the per-wait 5 ms is not the link - the same frames over adb forward move it by only a quarter, which is what the handoff measured and could not explain 2026-09-25 21:14:15 +08:00
BZLZHH f99415a493 [Docs] (P12): the atlas window's exact composition, and the negative result that kills the record-dedup idea before anyone builds it 2026-09-25 21:03:42 +08:00
BZLZHH 1d363eb109 [Perf] (MG_Remote): the client counts RECORDS SENT per op beside the waits each row PAID, and the pair says item B stopped the waiting without stopping the sending
- A wait count alone cannot say whether a frame is slow because it sends too much or because it
  waits too often. The device run made that the next question: one atlas frame paid 43,232 waits
  while the upload row paid none.
- LinkMetricsNoteRecord is called at the encoder's one commit point (PipeWireCodec.cpp, beside the
  existing CallClass::WireRecords count, for the comment's reason: after chunking no emitter knows
  how many records its call produced). LinkMetricsRecordsFor reads it back, and every frame that
  emitted something now logs `P65LinkMetrics kind=frame-sent frame=N records=M by_op=...`.
- The host measurement now prints and asserts both halves for the same workload. An atlas-shaped
  load (64 sub-uploads into one level) sends 128 ResourceSubData records and waits for ZERO of
  them, while ResourceCreate 76/76, ResourceRespecify 130/130 and SetTextureParams 130/130 are
  one wait per record. That is item B stated exactly: the records still go out; the round trips do
  not. It also says what the next thing to fix is - the record COUNT of those three rows, not their
  waits, which the contract rules they must keep.
2026-09-25 20:52:01 +08:00
BZLZHH d057e1cb8a [Docs] (P12): the first device run of A and B - the crash path walked twice without a crash, and the atlas frame says item B is necessary but not sufficient 2026-09-25 20:46:23 +08:00
BZLZHH 170b789e7f [Perf] (MG_Remote/Transport): the per-op reply-wait split prints with every frame, because a whole-session total cannot name the rows behind one frame
- The device run that motivated this made the point: 43,232 of a session's 43,629 reply waits were
  in ONE atlas frame, and the exit summary never appeared at all - a SIGTERM does not take the
  session's clean stop path, so LinkMetricsEnd is never reached. The split added one commit ago
  was therefore invisible on exactly the run that needed it.
- Emit now writes a second line for any frame that bought a wait:
  `P65LinkMetrics kind=frame-op frame=N wait_replies=M by_op=<id>=<count>,...`. The formatting is
  factored into FormatOpSplit and shared with the exit line. Frames that bought nothing stay
  quiet, so a steady-state session does not double its log.
- Measured on the device (MC 26.2 into a world over TCP, MOBILEGL_PIPE_STATS=1): the steady state
  is one wait per frame and it is op 9 (Present); the upload-heavy startup frames are
  `2=50,3=68,47=62` - ResourceCreate / ResourceRespecify / SetTextureParams, with ResourceSubData
  (48) absent, which is item B working on the wire it was written for.
2026-09-25 20:45:18 +08:00
BZLZHH e734880cf2 [Docs] (P12): the device page names the by_op ids, and the note says where the per-op line comes from 2026-09-25 20:20:24 +08:00
BZLZHH 618395529f [Perf] (MG_Remote/Transport): the per-op reply-wait split also lands in the exit summary, because a spawn client never prints a frame line
- The tally added beside this is only useful if somebody can read it where the problem is. The
  handoff says why that was hard: under spawn the client does not advance the frame window, so the
  per-frame P65LinkMetrics lines never print, and a SIGKILL takes the exit-time dump with it -
  "only the dump at exit, and SIGKILL does not get it".
- LinkMetricsEnd writes the split on its own line from the one place a client that ends the
  ordinary way always reaches: P65LinkMetrics kind=per-op wait_replies=N by_op=<id>=<count>,... .
  Ids stay raw (this file includes only <cstdint> and friends, not the op table) and the reader
  names them: 2 ResourceCreate, 3 ResourceRespecify, 47 SetTextureParams, 48 ResourceSubData. The
  mapping is written down in mgl-device/README.md, which is the page the device run reads.
2026-09-25 20:20:04 +08:00
BZLZHH ce0235b190 [Docs] (P12): the three rows that keep their wait get a ruling with evidence behind it, and the handoff's open op-split question gets its numbers
- What was open: the handoff's 4.3 called ResourceCreate and SetTextureParams "a second problem of
  the same kind, needing its own analysis" and left it there; 5.4 said the per-op split of the
  55,428 waits was never quantified. The measurement committed beside this answers both.
- The ruling (CONTRACT-P5E.md 2.5): ResourceCreate, ResourceRespecify and SetTextureParams stay
  kWaitReply, because each answer carries a fact about the SERVER's record table that the client's
  own tracker provably cannot know. RespecifyOnce says it outright - "the applier's REFUSAL is the
  only signal that says 'I hold nothing for this handle'", which is the scope
  MGPipeApplierReleaseObjectRecords leaves for a served context whose frontend objects live on;
  EmitTextureParams uses its own refusal as the same self-heal trigger; PublishCreate may not take
  the publication latch on a refused create (D-I1/c0b) or the death path emits a resource_destroy
  for a record that does not exist. Unlike the upload half these are not "the emission gate already
  answered it" rows, so removing their waits removes the signal.
- The numbers, both shapes a stitcher produces: an atlas stitch goes 64 -> 0 reply waits; a texture
  per sprite goes 384 -> 320 with ResourceSubData now zero and the remainder all three of those
  rows. So item B hits exactly the shape it was aimed at, and what is left is per-resource /
  per-call rather than per-chunk.
- notes/p12/DEVICELOST-AND-UPLOAD-WAITS.md gains 2.4 with the table and the ruling, and its "second
  problem of the same kind" entry is closed rather than carried; MEASUREMENTS.md records the two
  shapes beside the 1 -> 0 single-upload number.
2026-09-25 20:17:22 +08:00
BZLZHH fa8b500ba8 [Perf] (MG_Remote/Transport): the reply-wait metric gains a per-op tally, and the measurement says which rows actually buy the waits
- The handoff measured the total (55,428 reply waits in one Minecraft frame) and left "which rows"
  open: a total is the same number whether the waits are uploads, creates or parameter sets, and
  those three have different fixes - so knowing the split is what says whether P12 item B is the
  whole story or a fraction of one.
- LinkMetricsBeginReply now takes the op and Window carries a per-op tally, so
  LinkMetricsReplyWaitsFor(op) reads back what the P65LinkMetrics line prints as `wait_replies`
  split by row. Same activation as everything else in that file (MOBILEGL_PIPE_STATS=1), one array
  in the window, and a value past the table lands on the last slot so the per-op numbers still SUM
  to the total.
- RemoteClientControls.ReplyWaitsByOpForAnAtlasShapedLoad drives the two shapes a stitcher
  produces and prints the table: an atlas stitch (64 glTexSubImage2D into one level) goes from 64
  reply waits to ZERO, and a texture per sprite (64 textures, 256 glTexParameteri) from 384 to 320
  with the remainder all ResourceCreate 76 / ResourceRespecify 130 / SetTextureParams 130.
- It prints before it asserts: the table is the point of the case, and a gate that fired first
  would hide the numbers on exactly the run that needs them - the red one. The assertions are the
  structural ones that can be wrong about the mechanism (upload rows zero, the other rows not, the
  per-op tally summing to the total); the numbers themselves are recorded, not gated.
- The A/B is a real run: with the wire-arm arm test removed the case goes red with the full table,
  which is where the 64 and 384 above come from.
2026-09-25 20:17:13 +08:00
BZLZHH bfd6392668 [Docs] (P12): the P5E contract records the texture half's paid-off wait, and the round's note carries the measured G1
- CONTRACT-P5E.md 2.5 said "the TEXTURE half keeps its wait because DrainTextureSubData clears the
  level's dirty flag on the accepted reply (D-D5; trailing)". The trailing is paid, so the clause
  now states the three facts that make that answer DERIVABLE on the wire arm (the emission gate IS
  the consumer question the applier's own acceptance belt asks; what remains is a loud bug class;
  a client-cancelled record still answers "not accepted"), where the wait moves to (SEG_STAGE),
  that MGPipeSubDataWantsItsReply stays the one place that decides, and the gate with its red-once.
  ResourceCreate and SetTextureParams are named as analyzed separately rather than swept in.
- notes/p12/DEVICELOST-AND-UPLOAD-WAITS.md: 2.3 now carries the G1 A/B this machine ran - .text
  13872002 bytes on both sides, 31746 symbols, 0 added / 0 removed, no per-symbol difference -
  instead of deferring it to CI, and 3 trades its "G1 not run" item for the device tooling
  (mgl-device/: launch-mc.py, gl-aliases.sh, start-phone-server.sh) the next round needs.
2026-09-25 20:00:44 +08:00
BZLZHH 3b9f536dd2 [Docs] (P12): the device-lost reply tolerance and the texture-upload wait granularity get their own record - what landed, the gates that can go red, and what is still owed
- notes/p12/DEVICELOST-AND-UPLOAD-WAITS.md: the two items, their root causes with file:line anchors,
  the five cases that pin them, the red each one produced when the fix was taken back out, and the
  four things this round did NOT do (device end-to-end, local G1, the ResourceCreate /
  SetTextureParams pair, and the sandbox's missing /dev/dri - which reds the two EventForfeitPeer
  lanes for reasons that have nothing to do with the change).
- p12/README.md: a line in the summary and the new file in this directory's table.
- MEASUREMENTS.md: the P12 conclusion and row now carry the number the gate pins - one reply wait
  per pure texture upload becomes zero - with the device wall clock named as the measurement still
  owed rather than left implied.
2026-09-25 19:44:51 +08:00
BZLZHH 33e02f4bcf [Perf] (MG_Pipe, MG_Remote/Client): on the wire arm a texture upload stops owning a reply - the wait moves from every record to the SEG_STAGE window
- MGPipeSubDataWantsItsReply is the ONE place that decides whether a resource_subdata record waits
  for its answer (WireTables.cpp:370 and PipeRoute.h:328 both read it, and the comment beside it
  says that a second spelling is how the client comes to wait for an answer the emitter told the
  server not to bother with). The buffer half has answered "no" since P5e; the texture half now
  does too, ON THE WIRE ARM ONLY.
- THE ANSWER IS DERIVABLE THERE, which is the argument - not that nobody reads it:
  * The emission is already gated on the answer's content. A texture record is produced only where
    FamilyIsLive(kMGPipeSubsystemTextureResources, ...) holds, and under split that gate IS
    CapsMirrorInstance().ServerConsumes(bit 10) (PipeFill.cpp:1719-1740) - the same published fact
    the applier's own acceptance belt reads (PipeApply.cpp:1408-1419, NoP4aConsumer). So the one
    refusal this acceptance exists to carry, "no consumer" (ID-39 / D-D5 / ID-18 M3, which is why
    the texture half ever asked), cannot be the answer to a record this client emitted.
  * What remains is a bug class, and it is loud already: a record that names no texture, declares
    no texels or carries no bytes is an MGP_TRIP_WIRE_REPORT (PipeApply.cpp:1201-1218, :2245-2252),
    and a handle the applier does not hold is counted and named (ResolveResourceIn, :721-728) with
    the client's own tracker preventing it from arising (PipeFill.cpp:1045-1051).
  * Every answer the client can still get for free, it still gets: the emitter reads this route's
    Bool exactly as before, so a record the CLIENT cancelled (the device-lost latch, an exhausted
    stage) still answers "not accepted" and the level stays dirty and is retried - D-D5's safe
    direction, unchanged.
- THE WAIT MOVES TO THE STAGE. A run that does not fit SEG_STAGE reclaims retired bytes and parks
  on the stage doorbell (PipeWireCodec.cpp StageAllocate / ReclaimStagedBytes) - the one wait a
  run-ahead client owes, and it is proportional to the WINDOW rather than to the call count. On
  Minecraft 26.2's atlas frame that is the handoff's 55,428 reply waits at ~4.4 ms each: 96 MiB of
  uploads through a 32 MiB window.
- THE GATE IS THE NUMBER ITSELF. LinkMetricsReplyWaits is added so a case can read back the same
  `wait_replies` the per-frame P65LinkMetrics line prints - LinkMetrics.cpp:76-80 counts exactly
  the records that own a reply slot. TextureUploadsStopOwningAReplyOnTheWireArm asserts that a pure
  glTexSubImage2D crosses the wire and moves it by ZERO, while a resource_create row - kWaitReply,
  and it always was - still moves it, so the zero cannot be "nothing was counted". Red-once run:
  with the arm test removed the child exits 142.
- Monolith and every non-wire arm are untouched, and that is pinned twice: the existing
  RemoteRunAhead.OnlyTheTextureHalfOfResourceSubDataWantsItsReply still reads true with no arm
  installed, and the new test sits inside MOBILEGL_BUILD_DISAGGREGATED with the original expression
  as its fall-through, so a pull build preprocesses to the line it had (G1).
- Verification debt, stated rather than implied: the host gate is the COUNT, not a wall clock.
  In-process control sessions run lockstep (the caps peer publishes no kCapRunAheadApply), so the
  verb barrier still waits there; the end-to-end wall-clock win is the device measurement the
  handoff describes and it has not been taken yet.
2026-09-25 19:44:38 +08:00
BZLZHH 096fa1a8c1 [Fix] (MG_Remote/Client): a DECLINED reply is an answer, not a corruption - the fence and query consumers stop folding status 1 into the abort, and the uncarried-bytes self-check spares a session that has latched device-lost
- ReadFenceReply (EmitTables.cpp) and ReadQueryReply (QueryEmit.inc) tested `status != kStatusOk`,
  which is two values in one branch. ReplySink's own header says "DECLINED IS A REAL ANSWER, not
  a failure" (PipeWireCodec.h:457-460), and the escape rows are held to the other half of the same
  rule - "ERROR IS NOT A DECLINE ... an escape may not fold 2 into `false`" (WireTables.cpp:446-450).
- BOTH VALUES OF 1 ARE REACHABLE WITHOUT A BYTE OF CORRUPTION: the server declines these five rows
  by design whenever it cannot answer them (PipeWireCodec.cpp:1820-1830 for the two fence rows,
  :1841-1861 for the three query rows - a zero-length answer to a kWaitReply row, R-5), and the
  client produces the same status itself once the session stops carrying records
  (ClientSession.cpp:1837 the device-lost latch, :2025 a doorbell that died during teardown, where
  the comment says outright that "the verb did not happen" is reported as DECLINED and not ERROR).
- WHAT THE FOLD COST: after a clean device loss Minecraft 26.2's next glClientWaitSync - the
  chunk-upload fences - died of Fatal{ProtocolCorruption, "Fence.reply"} (rc=134) instead of taking
  the no-op the design promises ("GL calls become no-ops", design/08-runtime-and-platform.md:33).
  The query rows had the same shape waiting for them.
- The no-op is the one this frontend already gives for a sync it cannot wait on: GL_Sync.cpp:103-106,
  PipeApplier.cpp:175-177 and DirectGLES.cpp:16489 all answer GL_ALREADY_SIGNALED. A query row
  answers its own "no answer yet" value - availability false, Produced 0 (which the caller already
  returns as false with the out-parameter untouched), timestamp 0.
- ERROR and a short OK keep the named Fatal they always had: a transport fault is not the server
  saying no, and an OK answer that arrived empty is a reply nobody can read. ReadPixels keeps its
  deliberate Fatal{ReadbackDeclined}: unlike these rows a blocking readback cannot succeed short of
  complete pixels, and that ruling is untouched.
- PipeFill.cpp's UncarriedInitialBytes self-check now spares a session that has latched device-lost.
  The resource_subdata it counts is DECLINED by design there, so "it emitted nothing" is the latch
  working rather than bytes this client forgot to carry - and there is no server left for them to
  exist on. P12's exit gate (a) is that killing the server gives a CLEAN device-lost latch, not a
  corpse at the next glBufferData that carries data.
- Four cases in RemoteClientControls.inc drive the production consumers over a live session whose
  peer declines them. Red-once run for all three Fatals (Fence.reply, Query.reply,
  UncarriedInitialBytes), and ErrorFenceReplyStillRefusesByName is the control that keeps the
  tolerance from becoming blanket. RemoteClientTest 166/166; the host unit lane 2610/2610.
- Split only: both consumer files compile under MOBILEGL_BUILD_DISAGGREGATED and the PipeFill hunk
  sits inside its `#if`, so a pull build preprocesses to exactly what it had (G1).
2026-09-25 19:43:40 +08:00
BZLZHH c984d95205 [Fix] (MG_Remote): CapsSnapshot.apiVersion carries the backend's own API version string, so a split client's GL_RENDERER names the real device instead of the target GL version (split build only; the pull build is unchanged, G1)
- PublishCapsSnapshot filled that field with info.RendererGLInfo.TargetGLVersion.toString(). The
  client reads it back through BackendObject_Remote::GetBackendAPIVersionString (CapsMirror's
  ApiVersion) and prints it in the parentheses of GL_RENDERER, which is the only place a real
  device is named; a monolith build fills the same parentheses from the backend's own
  GetBackendAPIVersionString. A split client therefore answered "Espryt (MobileGL Core) (4.6.0)"
  where monolith answered "Espryt (MobileGL Core) (Adreno (TM) 750, OpenGL ES 3.2)".
- The target version still crosses the wire inside the renderer-info blob (CapsCodec), which is
  where GL_VERSION reads it, so no reader loses it.
- Content only: the field already existed and the blob layout is untouched, so the wire
  fingerprint needs no bump, and the file compiles only under MOBILEGL_BUILD_DISAGGREGATED.
- Verified with a PC client against a phone server: Minecraft and Sodium both log
  "Espryt (MobileGL Core) (Adreno (TM) 750, OpenGL ES 3.2)".
2026-09-25 17:25:21 +08:00
swung0x48 41b35536e3 [Docs] (P12): record post-review verification 2026-09-25 01:34:29 -04:00
swung0x48 0fe015886e [Docs] (Disaggregated): the top-level files become short indexes in plain language; the design text moves to design/, the how-to and reference material to guide/
- README, ARCHITECTURE, ROADMAP, MEASUREMENTS and CURRENT_STAGE_PROGRESS are rewritten as indexes
  (84 KB -> 15 KB): what the project is, where it stands, and where each detail lives. ROADMAP is one
  plain sentence per phase; MEASUREMENTS is the five conclusions that matter plus one line per phase;
  CURRENT_STAGE_PROGRESS is the P12 goal, a checklist and the next steps.
- ARCHITECTURE's body moves unchanged into ten design/ files, keeping §1-§17 and appendices A/B, so
  `ARCHITECTURE.md §N` references in code resolve through the table in ARCHITECTURE.md.
- guide/ holds build-and-run, code map, glossary and the engineering discipline (from README and
  ROADMAP), and the device pinning record that used to sit in devices/.
- The open-debt table moves to notes/DEBTS.md; the detailed current-stage tables (with decision IDs,
  the open line items and evidence locations) move into notes/p12/README.md. notes/README.md maps
  every moved path and section.
- The documentation citation lint in test.yml also covers design/ and guide/, which describe the
  current tree as the top-level files did.
2026-09-24 22:39:46 +00:00
swung0x48 43b5cfa94e [Docs] (Disaggregated): the four index documents keep only summaries, and every phase's detail lives in its own notes/<phase>/README.md
- ROADMAP, MEASUREMENTS and ARCHITECTURE are rewritten as summaries (333 KB of index text -> 84 KB).
  ROADMAP's phase table is one line per phase plus a link; MEASUREMENTS keeps headline numbers under
  its old §1-§13 numbering (plus §14-§16 for P6.5, P7 and P12); ARCHITECTURE keeps every decided rule
  under the same §1-§17 / appendix numbering, so the `ARCHITECTURE.md §N` references in code still
  land, and §17 becomes an index of the phase landing shapes.
- CURRENT_STAGE_PROGRESS now describes the current stage (the P12 on-screen-server subset) and absorbs
  CURRENT_STAGE_STATUS, which is removed. The P7 status tables, the P6 closing snapshot and the P6.5
  first-wave snapshot move to notes/p7, notes/p6 and notes/p65.
- Each phase gets notes/<phase>/README.md: a short summary, then the verbatim material that left
  the index documents (its ROADMAP row, its MEASUREMENTS section with subsection numbers kept, its
  ARCHITECTURE §17 part, milestones) and a file index. P8-P11 and P13 get their planned scope;
  the full open-question list moves to notes/OPEN-QUESTIONS.md.
- The phase reports that sat at the top level move into their phase folders with unchanged names:
  P5D-INPROC-PERFORMANCE, P5E-RUNAHEAD, P5F-WIRE-COMPLETENESS, P6-SPAWN-PLAN, P6-CONTRACT-DRAFT,
  P6-ENDSTATE-REVIEW. Their relative links, CONTRACT-P6.md's links to them and one tool docstring
  are updated; notes/README.md maps every old path and section number to its new place.
- 71 files under notes/recovered/ that were byte-identical copies of phase-folder files are removed.
- Corrections made while condensing: the server process reads Transport as Spawn (CONTRACT-P6 §3.1
  amended the old "child forces monolith" rule); both backends now publish kCapRunAheadApply;
  MOBILEGL_IPC_RESPAWN is a named refusal and MOBILEGL_IPC_IDLE_EXIT_S is not parsed; the Redmi
  pin row is in tools/device_bench/pin_device.sh.

check_doc_citations over docs/Disaggregated/*.md: 0 problems. No relative link is newly broken.
2026-09-24 22:21:09 +00:00
swung0x48 1fb18d9e73 [Docs] (P12 review): the status page says the P12 subset and its review round are integrated into feat/disaggregated, and that the device re-check is the next thing owed 2026-09-24 14:03:49 +00:00
swung0x48 341b43fa4f [Docs] (P12 review): ID-P12-15 - G1 holds for the review round: the pull build at cb0404f2 and at this head, same machine and compiler, has .text 0x9b1bf2 on both sides and 30614 symbols with 0 added / 0 removed 2026-09-24 14:03:48 +00:00
swung0x48 2cf3298023 [Docs] (P12 review): ID-P12-14 records the review round's host gate - the new cases' red-without-fix demonstrations, and every remaining red traced to the container's CMake/CTest, not to the fixes
- unit 2605/2605, integration-spawn/tcp and the three Magma two-process lanes green, hygiene 12/12, census 79 / ratchet 171 unchanged, revision 3 pinned.
- Major 1 red without its fix ("Detach answered TimedOut in 3000 ms after 200 of 200 streamed records"), major 2 red without its fix (all four expectations), both green restored.
- SplitLogPaths / spawn_lane_parity / integration-gpu reds are registration artifacts of this container's CMake 3.28 (lane environment missing) and 4.4 (lane environment flattened into one variable; gate-review-d04ac7d1-cmake4.log); the four PersistentCoherentMapScenario and three F1WireScenario cases behave identically at cb0404f2 and at this head when the binary is run directly with the same environment. No CMake file changed since cb0404f2.
2026-09-24 14:03:48 +00:00
swung0x48 b240bd942f [Docs] (P12 review): the review round's host gate log and the container variant of the gate script
gate/gate-review-3c342c97.log: the P12 gate at head 3c342c97 (dirty=0) in a 4-core container with clang-19, lavapipe (mesa 25.2) and ctest 4.4.3 - unit 2605/2605 (the four new review cases included), integration-spawn/tcp and the three Magma two-process lanes green, hygiene 12/12, fatal census 79 / link ratchet 171 unchanged, protocol revision 3 holds. Still being attributed: SplitLogPaths and spawn_lane_parity (the container's ctest tooling; no CMake file changed since cb0404f2), four PersistentCoherentMapScenario and three F1WireScenario cases in integration-gpu (being run at base cb0404f2 with the same flags), and IterationRPProgram203Scenario (out of scope, ID-P12-12). gate-container.sh is gate.sh with the host paths removed and G1 run separately.
2026-09-24 14:03:48 +00:00
swung0x48 b7ad6113ee [Docs] (P12 review): the P12 decision log, ARCHITECTURE §15.2's on-screen path, and the ROADMAP P12 row record the review round
- notes/p12/INTEGRATOR-DECISIONS-P12.md (new): ID-P12-1..4 the subset's shape (the user rulings, one surface mode per session, the lost-window order, the in-process thread hand-off), ID-P12-5..11 the review round's ten fixes (two majors: the lost window answered before every ring record, the ended session's Espryt unit shadows emptied with its twins), ID-P12-12 IterationRP out of scope, ID-P12-13 the device stage not re-run in this environment.
- ARCHITECTURE.md §15.2: the deleted Surface -> AIDL window path is struck through, and the server-owned window path, the lost-window contract and the in-process session boundary are described.
- ROADMAP.md: the P12 row's status names the review round and that the device checks were not repeated.
2026-09-24 14:03:48 +00:00
swung0x48 251d333231 [MG_Backend] (P12 review): Magma's renderer is dropped when a server session ends, so a session that ended without a clean teardown leaves nothing attached to the server's window
The Vulkan renderer (swapchain and VkSurfaceKHR on the window) is a process global that only ReleaseEGLResources or a released surface drops, and ~BackendObject_DirectVulkan was defaulted. A client that went away without eglTerminate (a crash, a kill, a dropped connection) sends neither, so after its session the renderer stayed on the display Activity's window past the end of the session's lease. Under a transport the server backend's destruction - the session's end, on the apply thread, before the lease ends - now drops the renderer (with the generation bump and the program reflection caches, as ReleaseEGLResources does). Split build only; the pull build's destructor is still defaulted (G1).
2026-09-24 14:03:48 +00:00
swung0x48 74ba390db5 [MG_Backend, MG_Test] (P12 review): the ended server session's texture and sampler unit shadows are emptied with its twins, so the next session's binds are not skipped against freed twins
Major. DropEveryTwinForEndedServerSession drops every twin with InProcessTeardown() answering true, which is exactly the case in which a texture / sampler twin's destructor does NOT scrub itself out of TextureImpl::g_boundTexturesCache / SamplerImpl::g_boundSamplersCache. The shadows kept raw pointers to freed twins; the next in-process session's twin allocated at a recycled address read "already bound" and skipped its glBindTexture / glBindSampler on a context where nothing was bound, so its uploads went to texture 0 and it rendered black. The drop now empties both shadows and puts the active-unit shadow back to 0 (the new context's GL_TEXTURE0). SanityTest EsprytServerSession.TheServerBackendsDestructionEmptiesTheTextureAndSamplerUnitShadows (split build only; the pull build is unchanged, G1).
2026-09-24 14:03:47 +00:00
swung0x48 ee456ccc40 [MG_Util, MG_Remote] (P12 review): the in-process display server forwards only its live session's own log lines to that session's client, and sends them outside the log mutex
SetLogForwarder forwards every server-role thread's line, which is right for a session child (the process is the session) and wrong for the display Activity's process, where MOBILEGL_IPC_ROLE=server makes every thread a server thread: the UI thread's surface callbacks and the listener's lines - naming other peers' addresses, Busy refusals included - went into the live client's control socket. And the forward was a blocking SendFrame under the global log mutex, so a client that stopped reading held that mutex and hung the UI thread at its next log line (surfaceDestroyed's Detach logs).

In-process, RunSession now installs SetSessionLogForwarder: only the session thread and the threads that opt in (the session's apply thread, SetThreadForwardsLogToPeer) forward, and the send happens after the log mutex is released, under a forward mutex of its own (WithLogBarrier takes both, so the flush ack still follows every send that began; a line logged by the send itself is not re-sent). The forked session child keeps SetLogForwarder unchanged; the pull build's Log() is unchanged (G1).
2026-09-24 14:03:47 +00:00
swung0x48 7680b78621 [MG_Backend] (P12 review): an ended server session's swap interval is forgotten, so the next session in the in-process display server starts at the driver's default
DirectGLES keeps the requested eglSwapInterval in a process global on purpose (a surface re-created mid-session re-applies it), and every server process used to serve one session. The in-process display server serves the next session in the same process, which then silently inherited the previous client's interval. The server backend's destruction under a transport now calls DirectGLES::ForgetRequestedSwapInterval() beside DropEveryTwinForEndedServerSession (split build only; the pull build is unchanged, G1).
2026-09-24 14:03:47 +00:00
swung0x48 29a82617e0 [MG_Remote, MG_Backend] (P12 review): a server-owned surface resizes the server's window and reports its real size, every server-owned surface keeps its geometry, and Espryt reports a window's extent only for the server's own window
- Client resize: eglResize on a headless client's server-owned surface only rewrote the backend's record of the window - the phone's SurfaceView kept its size - and the client's EGL state answered the size it asked for. The server now recognises the surface as its own (ServerLoop::m_serverOwnedSurfaces, per session) and resizes it as a geometry request (the lease re-taken by its holder, the display waited for), resizes the backend surface to the window's real extent and replies with it; BackendObject_Remote adopts that extent (ServerResizeServerOwnedWindowSurface).
- Size tracking: the client held ONE server-owned surface, so a second one took the first's geometry updates and releasing it left the first with none. ClientSession now tracks every server-owned surface and applies an extent to all of them (they share the server's one window).
- Size reports without the knob: Espryt's new extent publish fired for every window surface of a spawn/tcp session, including windows a client named itself without MOBILEGL_IPC_SURFACE=server, switching their consumer to the reallocating arm. ServerLoop now names its own leased window to the backend (MGPipeServerSetOwnedWindow, split-only) and Espryt publishes the extent only for that window.
2026-09-24 14:03:47 +00:00
swung0x48 277940f39f [MG_Remote, MG_Test] (P12 review): a session asking for the window's own size after a fixed-size session waits for the layout's size instead of leasing the stale fixed one
AcquireFor(0, 0) granted the lease at once, at whatever the window reported - and right after a session that fixed the size, that is still the fixed size: setSizeFromLayout answers asynchronously, like setFixedSize. ServerDisplay now remembers the geometry last requested and the layout's last reported extent; a 0/0 request that follows a fixed one waits (within the geometry grace) for the layout's extent, or for any report after the request that is not the old fixed size. ServerDisplayTest.TheWindowsOwnSizeAfterAFixedSizeIsTheLayoutsNotTheStaleFixedOne.
2026-09-24 14:03:46 +00:00
swung0x48 8f2a084354 [MG_Remote, MG_Test] (P12 review): a lost server window is answered before every ring record, and a stopping in-process server leaves a streaming client's queue unapplied
- Major: the window-lost release was picked up only between drain batches (PumpControlRequest), and a client streaming frames keeps the command ring from emptying for seconds - surfaceDestroyed waited out its 3 s bound on the UI thread, answered TimedOut, and the session kept presenting into the destroyed window. DrainRing now checks the request after its latch check, before every pop. ServerLoopLatchTest.ALostServerWindowIsReleasedMidBatchWhileTheClientStreams streams a 4 s batch under Detach.
- Stop while streaming: mobilegl_server_stop_inprocess ended the session through loop.Stop(), whose batch and final drain of a ring the client keeps refilling could outlast the bounded join (Fatal{ApplyThreadJoinTimeout} aborts the display Activity's process). RunSession now calls ServerLoop::AbandonQueuedRecords() first when the server is stopping: the drain stops after the record in hand and the exit path declines the rest, as a ReverseChannelForfeit does. ServerLoopTest.StoppingTheServerUnderAStreamingClientLeavesTheQueueAndJoinsPromptly.
2026-09-24 14:03:46 +00:00
swung0x48 cb0404f2b9 [Docs] (P12): the on-screen server window's plan, audit, harness and device evidence move into the tree, and the status page says what this line is and what it still owes - CURRENT_STAGE_STATUS records that the P12 subset is implemented but NOT handed off (both exit gates unrun: no cross-machine TCP, no FCL + kill-server device-lost), the ROADMAP P12 row stops reading "待排", and notes/p12 carries the brief, the four read-only audits, the gate and device harness, and the curated device evidence (check transcripts, replay verdicts, on-screen captures; logcat left out per notes/README.md's evidence-data rule) 2026-09-24 07:44:40 -04:00
swung0x48 29b7b284b9 [tools] (P12): trace_replay README - the two device servers, tcp_device_server.py --surface window, and the headless client's MOBILEGL_IPC_SURFACE=server --window-surface replay 2026-09-24 00:51:13 -04:00
swung0x48 e7486f8a4c [android-plugin] (P12): MobileGLServerService logs its supervisor's closed output as the service stopping, not as a failure with a stack trace, when the display Activity stops it 2026-09-24 00:51:13 -04:00
swung0x48 06dad59dbd [MG_Remote] (P12): the in-process pin line names the pin it records - MOBILEGL_BACKEND_TYPE when the environment set it, the first session's otherwise 2026-09-24 00:51:13 -04:00
swung0x48 a4a940db2b [MG_Remote] (P12): the display geometry up-call attaches the apply thread to the JVM under its own name, so the rest of the session is not logged as the up-call's thread 2026-09-24 00:40:35 -04:00
swung0x48 c2758abdf9 [MG_Backend, MG_Test] (P12): the server backend's destruction drops every Espryt twin under a transport, so the in-process display server's next session no longer adopts the ended session's programs, VAOs and buffers (split build only; the pull build is unchanged, G1) 2026-09-24 00:40:35 -04:00
swung0x48 984db45576 [tools, scripts] (P12): tcp_device_server.py --surface window|pbuffer (default pbuffer, unchanged) starts the on-screen display Activity on a lit screen, and --backend pins either server 2026-09-24 00:20:11 -04:00
swung0x48 3726b1f7e5 [android-plugin, tools] (P12): MobileGLDisplayActivity - the on-screen display server in :mglwin (software window, fullscreen SurfaceView, aspect-fit geometry, env applied before the library loads), one shared server environment with MobileGLServerService, and one server at a time (the Activity stops the service, the service kills :mglwin) 2026-09-24 00:20:11 -04:00
swung0x48 02801c2acb [MG_Remote] (P12): the in-process display server's JNI - install the display with ANativeWindow hooks and a JVM-attached geometry up-call, serve/stop on a Java thread, attach from surfaceCreated/Changed and a blocking detach from surfaceDestroyed (Android split build only) 2026-09-24 00:20:11 -04:00
swung0x48 37b53152cc [MG_Test] (P12): host proofs - the in-process server entry on a thread (sequential sessions, latch reset, Busy, backend pin, stop), a forked session child's parent-death signal, and a MOBILEGL_IPC_SURFACE=server client refused by name on both sides by a display-less server 2026-09-23 23:55:23 -04:00
swung0x48 4fb5aadfa2 [tools] (P12): Linux glws creates a headless window surface for MOBILEGL_TRACE_SURFACE=window - the MobileGL server's window under MOBILEGL_IPC_SURFACE=server, refused by name otherwise 2026-09-23 23:55:23 -04:00
swung0x48 2fa4714fc9 [MG_Backend] (P12): Espryt publishes a remote session's window-surface extent with its default-framebuffer format (split build only; the pull build is unchanged, G1) 2026-09-23 23:55:23 -04:00
swung0x48 826584d265 [MG_Remote] (P12): the in-process display server - RunSession returns its exit code, TcpSupervisor hands authenticated sessions to threads (Busy, ResetSessionLatch between sessions, the backend pinned for the process, a non-blocking stop), and forked session children die with their supervisor (PR_SET_PDEATHSIG) 2026-09-23 23:55:23 -04:00
swung0x48 d481393f8f [MG_Remote, MG_Impl, MG_State, Config] (P12): MOBILEGL_IPC_SURFACE=server - headless window surfaces (NULL accepted), one ServerOwned CreateWindowSurface and no SetWindowHandle, the server's geometry into the EGL state before the create returns, and every refusal named on the client 2026-09-23 23:55:23 -04:00
swung0x48 844b217661 [MG_Remote, MG_Test, scripts] (P12): the server's own window - ServerDisplay (an opaque window with hooks, a lease, a bounded Detach), the ServerOwned arm with the one-mode-per-session latch and the named no-display refusal, the lost-window release on the apply thread and the ServerWindowLost family 2026-09-23 23:55:23 -04:00
swung0x48 457aebe7f0 [MG_Remote, MG_Test, scripts] (P12): WindowKind::ServerOwned on the wire - control revision 3 with SurfaceRefusal and SurfaceReply width/height/refusal, the codec's ServerOwned rules (creation only, token 0, SetWindowHandle a named refusal) on a frame-local marker, and the non-zero-token latch row 2026-09-23 23:55:23 -04:00
swung0x48 9f669e5215 [docs] (P7): ID-P7-63 - P7 closes: the final device window reads gate 3 36/36 and gate 5 PASS, and the handoff records the end state 2026-09-23 20:51:30 -04:00
swung0x48 78e71f2b01 [docs] (P7): ID-P7-60/61/62 - the codex closeout fixes, the spawn cold-start heartbeat, and gate 3's nondeterministic-monolith ruling; window 2 reads gate 5 PASS 2026-09-23 19:34:55 -04:00
swung0x48 990bc649a6 [docs] (P7): CONTRACT-P7 7.2 and the window-2 RUNBOOK record the ID-P7-62 final ruling (1.5 x Delta_mm + 8, monolith x5), its simulated pass rates, gate 3's x5 timing and the window-2 re-reductions at 36/36 2026-09-23 19:34:03 -04:00
swung0x48 5662901100 [tools] (P7): ID-P7-62 final ruling: clause (2) bounds the nearest-monolith delta at 1.5 x Delta_mm + 8, and 30-gate3.sh runs the monolith x5 whatever --repeat 2026-09-23 19:34:03 -04:00
swung0x48 b44a1c3df3 [docs] (P7): CONTRACT-P7 7.2 and the window-2 RUNBOOK record the nearest-monolith form of ID-P7-62, its window-2 re-reductions and the simulated monolith x3 pass rates 2026-09-23 19:34:03 -04:00
swung0x48 b98dc9f890 [tools] (P7): ID-P7-62 clause (2) holds every split pass to its nearest monolith pass within 1.25 x D_mm + 32 px and Delta_mm + 1, and 30-gate3.sh / window.sh are executable again 2026-09-23 19:34:03 -04:00
swung0x48 2fafe1f438 [docs] (P7): CONTRACT-P7 7.2 records the ID-P7-62 nondeterministic-monolith amendment, and the window-2 RUNBOOK its ruling, re-reductions and bliss's pass rate under monolith x3 2026-09-23 19:34:03 -04:00
swung0x48 9d026e1ab6 [tools] (P7): gate 3 runs the monolith x3 and 60-reduce.py judges a case whose same-session monolith is not bit-identical by the ID-P7-62 distributional check 2026-09-23 19:34:03 -04:00
swung0x48 74636c8432 [MG_Backend, MG_Test] (P7): the wire arm's shader colour blit samples an unscaled blit through its nearest sampler, so a 1:1 GL_LINEAR glBlitFramebuffer into the default framebuffer is an exact copy 2026-09-23 18:59:20 -04:00
swung0x48 cfe830cf44 [MG_Remote, MG_Test, MG_IntegrationTest, scripts] (P7): a spawned server reports SurfaceProgress while its apply thread runs a surface op, so the client's reply budget bounds silence and a cold bring-up past 20 s is waited for 2026-09-23 18:17:02 -04:00
swung0x48 dd415408dd [docs] (P7): CONTRACT-P7 §10.7 and §12 and the ROADMAP Ph row record the int3 follow-ups: the narrowed latch window, null-or-unit-private image units and five new debts 2026-09-23 17:49:47 -04:00
swung0x48 2069887874 [tools] (P7): the window-2 reducer takes an adjudicated lost CTS case out of its block's rate on both sides and prints it ADJUDICATED 2026-09-23 17:48:25 -04:00
swung0x48 a23ae93753 [MG_Remote, MG_Test, scripts] (P7): the latch comments say the check before every pop narrows the control-thread window to check-to-pop and the session ends at the next check 2026-09-23 17:45:57 -04:00
swung0x48 9eb798a60e [MG_Backend, MG_Test, MG_IntegrationTest, scripts] (P7): a private image placeholder is keyed by (shape, unit) and capped, and the depth-resolve identity is built by a tested helper 2026-09-23 17:45:57 -04:00
swung0x48 8f3d179ced [tools] (P7): window-2 reducers enforce the contract's repeats, arms and case set, and gate Pass-to-NotSupported losses 2026-09-23 17:04:14 -04:00
swung0x48 02db0e4d9d [scripts] (P7): spawn_lane_parity names the two ImageUnitPrivate knob entries by one tail each 2026-09-23 17:04:14 -04:00
swung0x48 e1b47dc2f6 [MG_Backend, MG_Test] (P7): the depth/stencil resolve probe's verdict is memoized per device identity (vendor, device, driver version, pipeline-cache UUID), not once per process 2026-09-23 17:04:13 -04:00
swung0x48 a7848f0a03 [MG_Backend, MG_IntegrationTest, scripts] (P7): an invalid or empty image unit on the wire arm binds a null storage descriptor (VK_EXT_robustness2 nullDescriptor), else a placeholder private to its (binding, unit) 2026-09-23 17:04:13 -04:00
swung0x48 03bcf5e7fb [MG_Remote, MG_Test, scripts] (P7): DrainRing checks the session latch immediately before every pop, so a latch the control thread stores between two records stops the next one 2026-09-23 17:04:13 -04:00
swung0x48 1052b13a1a [MG_State, MG_Test] (P7): PH-5 bounds a vector count by its element's minimum encoded size, so compact archives the codec writes decode again 2026-09-23 17:04:13 -04:00
swung0x48 6afa077f96 [docs] (P7): ID-P7-57/58/59 - F2 finishes the Ph items, gate 5's three inproc CTS defects are fixed, and device window 2 is tooled 2026-09-23 15:52:33 -04:00
swung0x48 431bda67dc [MG_Backend, MG_IntegrationTest] (P7): a real-probe negative control proves the depth-resolve measurement can see a broken render-pass arm on the host 2026-09-23 15:20:17 -04:00
swung0x48 33f3133920 [docs] (P7): record g5-msprobe, the device-probed depth/stencil resolve arm order, with its evidence and closed debts 2026-09-23 15:13:45 -04:00
swung0x48 564602b71d [MG_Util, MG_Test] (P7): list the no-draw depth/stencil render-pass resolve in the Vulkan POST's Known Driver Bugs as FIXED 2026-09-23 15:13:45 -04:00
swung0x48 e83581c3c4 [MG_Backend, MG_Test, MG_IntegrationTest, scripts] (P7): choose the multisample depth/stencil resolve arm by a device probe with a shader-arm control instead of the Qualcomm vendor id 2026-09-23 15:13:44 -04:00
swung0x48 f7f0b41847 [docs] (P7): record g5-msrbo root cause, evidence and debts 2026-09-23 15:13:44 -04:00
swung0x48 fca8e9b21e [MG_Backend] (P7): widen wire image barriers to both aspects for D16_UNORM_S8_UINT 2026-09-23 15:13:44 -04:00
swung0x48 23be55e477 [MG_Backend, MG_IntegrationTest, scripts] (P7): pin the Qualcomm depth/stencil resolve arm order on the split and spawn arms 2026-09-23 15:13:44 -04:00
swung0x48 53397e6e1e [MG_Backend, MG_Test] (P7): resolve multisample depth/stencil with the shader arm first on Qualcomm 2026-09-23 15:13:44 -04:00
swung0x48 b1f9ea39f1 [MG_Backend, MG_IntegrationTest] (P7): copy 1:1 single-sample depth/stencil blits on the Magma wire arm 2026-09-23 15:13:44 -04:00
swung0x48 32cc68fa59 [tools, docs] (P7): window 2 holds the shared device lock and marks a CTS block done only when it is a complete reading 2026-09-23 14:39:40 -04:00
swung0x48 5a844b1e50 [tools] (P7): window-2 reducer never passes an empty, unrateable or subset reading, and its rate gate is tested 2026-09-23 14:39:40 -04:00
swung0x48 6b8690f1d5 [tools] (P7): window-2 reducers follow the contract formula, require every block and one boot session 2026-09-23 14:39:40 -04:00
swung0x48 cf54962a61 [tools, docs] (P7): window 2 restore stops the replay leftovers and rides out the vendor boost 2026-09-23 14:39:40 -04:00
swung0x48 09e98055b6 [tools, docs] (P7): device window 2 runbook and scripts 2026-09-23 14:39:40 -04:00
swung0x48 e663dec79e [trace_replay] (P7): the device runner starts from a POSIX host and a dead repeat archives no stale picture 2026-09-23 14:39:39 -04:00
swung0x48 941bd4be6a [MG_IntegrationTest] (P7): the texture-view window case joins the gated image-unit arms 2026-09-23 14:39:35 -04:00
swung0x48 3d5160d397 [MG_Backend, MG_IntegrationTest] (P7): bind the storage placeholder for an image unit naming a level or layer its texture lacks 2026-09-23 14:37:13 -04:00
swung0x48 0f64af8f55 [MG_Remote, MG_Test] (P7): state banded reads against PH-3's tight-answer bound, not DstSize equality 2026-09-23 14:37:08 -04:00
swung0x48 8af60919ff [MG_IntegrationTest] (P7): banded readback scenarios on every split arm and Magma 2026-09-23 14:37:08 -04:00
swung0x48 3874b547a9 [MG_Remote, MG_Test] (P7): band ReadPixels larger than one reply slot 2026-09-23 14:37:08 -04:00
swung0x48 3f6cadc4d4 [MG_Test] (P7): EventForfeitPeerTest pins the harness's headless EGL for the servers it starts, so both forfeit entries run on a runner with no DISPLAY 2026-09-23 13:55:54 -04:00
swung0x48 20e3d4f515 [MG_IntegrationTest] (P7): TcpLane.EventForfeitPeer takes the mobilegl-tcp lock the CI lock guard requires of every integration-tcp entry 2026-09-23 13:50:05 -04:00
swung0x48 f0a6d6f443 [docs] (P7): qualify the PLAN OQ-10 row's Init.cpp citation so check_doc_citations --strict resolves it 2026-09-23 13:45:47 -04:00
swung0x48 20562d33a3 [MG_Remote, docs] (P7): keep DrainRing's per-record latch check where ph_latch_sites pins it beside PH-6's forfeit check, and record the merged 100 ms control slice 2026-09-23 13:45:47 -04:00
swung0x48 13f4482830 [docs] (P7): record the f2-auth pre-auth knobs, the closed exposures and the debts left 2026-09-23 13:45:23 -04:00
swung0x48 efc899a1b0 [MG_Remote, MG_Test, scripts] (P7): bound the pre-auth cap and pause accepting when out of descriptors 2026-09-23 13:45:00 -04:00
swung0x48 bc62b17b0c [MG_Remote, scripts] (P7): refuse queued DataBinds on every pre-Accept exit and route DataBinds without a reap wait 2026-09-23 13:45:00 -04:00
swung0x48 5411180288 [MG_Remote, scripts] (P7): name every single-session first-frame outcome and pin its token-first order 2026-09-23 13:45:00 -04:00
swung0x48 04229e62d4 [MG_Remote, MG_Test, scripts] (P7): share the pre-auth queue between addresses and back off silent slot holders 2026-09-23 13:45:00 -04:00
swung0x48 6628674d72 [MG_Remote, scripts] (P7): refuse DataBinds queued on the session hand-off before it closes 2026-09-23 13:45:00 -04:00
swung0x48 3d7cc9592b [MG_Remote, MG_IntegrationTest, CI, scripts] (P7): add Ph fuzz arm 1 for malformed control frames 2026-09-23 13:45:00 -04:00
swung0x48 7bb1da73d4 [MG_Remote, MG_Test, scripts] (P7): authenticate TCP peers before fork with bounded pre-auth work and backoff 2026-09-23 13:44:38 -04:00
swung0x48 7aee3e2105 [ci] (P7): ph_latch_sites --table names each mechanics control by its gtest suite, not its file 2026-09-23 13:44:37 -04:00
swung0x48 1f67123835 [docs] (P7): record the PH-1 (3)(4) latch as CONTRACT-P7 §10 item 7 (session children latch and exit 75, inproc keeps Fatal) and its debts in §12; mark the PLAN PH-1, PH-2 and fuzz-arm-2 rows 2026-09-23 13:44:37 -04:00
swung0x48 3ef5c61e96 [MG_Backend, MG_Test] (P7): a PH-5 unit control that discriminates the element charge, @Ph-declined on PH-2's eight Fatal sites, and the mip-shadow comment made true 2026-09-23 13:44:37 -04:00
swung0x48 8848ee6ccf [MG_Remote, MG_Test, ci] (P7): red-once controls for the latch's decline-and-close half, a control pump that declines after the latch, the pre-gate ahead of the applier's stamp, and a PH-5 raw-peer row that discriminates 2026-09-23 13:44:37 -04:00
swung0x48 ec3758d088 [MG_Remote, MG_Test, ci] (P7): one raw-peer row per latch site and the D11 bounds against a --serve supervisor, each followed by a rendering next session (PH-1 (4), fuzz arm 2) 2026-09-23 13:44:37 -04:00
swung0x48 d681d9b1bf [MG_Remote, MG_Test, ci] (P7): peer-byte faults latch the session child and decline the rest; inproc keeps Fatal (PH-1 (3)) 2026-09-23 13:44:37 -04:00
swung0x48 caaedd7abe [MG_Remote, MG_Pipe, MG_Backend, MG_Test] (P7): repair F2 ID-49, byteless-level and R8 lane regressions 2026-09-23 13:44:37 -04:00
swung0x48 10c5168be2 [MG_Pipe, MG_Remote, MG_Test, MG_IntegrationTest] (P7): accept empty texture levels in the respecify extent carrier 2026-09-23 13:44:37 -04:00
swung0x48 38ac5dbd2d [MG_Remote, docs] (P7): record PH-6's debts in CONTRACT-P7 and fix the stale event-ring knob text 2026-09-23 13:44:37 -04:00
swung0x48 1f9ae2cfb5 [MG_Test] (P7): keep fuzz arm 3 from leaking its supervisor on a red run 2026-09-23 13:44:14 -04:00
swung0x48 96ab12b5ab [MG_Remote, MG_Test] (P7): end a forfeited session whose peer keeps talking on control 2026-09-23 13:44:13 -04:00
swung0x48 db4943601f [MG_Remote, MG_Test] (P7): pin the ended-control-stream note with a half-close case in fuzz arm 3 2026-09-23 13:44:13 -04:00
swung0x48 e259d2017a [MG_Remote, MG_Test, MG_IntegrationTest] (P7): end the PH-6 event wait on a stop or a dead shm peer, and read the wait knob back 2026-09-23 13:44:13 -04:00
swung0x48 1ef27dcd5c [MG_Test, MG_IntegrationTest, CI] (P7): add fuzz arm 3 - a peer that stops draining, on shm and tcp 2026-09-23 13:44:13 -04:00
swung0x48 3d0f531170 [MG_Remote, MG_Test, docs] (P7): forfeit the reverse channel on a non-draining peer (PH-6) 2026-09-23 13:44:13 -04:00
swung0x48 0535c4f07d [MG_Pipe, MG_Remote, MG_Backend] (P7): bind staged bytes to declared mip extents 2026-09-23 13:43:44 -04:00
swung0x48 501bdb2fed [MG_State, MG_Test] (P7): bound program archive vector counts 2026-09-23 13:43:44 -04:00
swung0x48 2bcec27ac2 [MG_Remote, MG_Test] (P7): bound readback scratch by server reply limit 2026-09-23 13:43:44 -04:00
swung0x48 54a8e3fadb [MG_Pipe, MG_Remote, MG_Test] (P7): bound create-render-state CSO slots 2026-09-23 13:43:43 -04:00
swung0x48 1997fd8878 [docs] (P7): ID-P7-55/56 - the CI reds since P6.5 are fixed and gate 4 closes with the p7-magma remainder tagged P13 2026-09-23 13:37:14 -04:00
swung0x48 d74e9923a7 [docs] (P7): record ratchet88 in CONTRACT-P7 4.1 and the 86 P13 symbols in 12 2026-09-23 13:23:09 -04:00
swung0x48 d49df3549f [MG_Backend, MG_State, scripts] (P7): retire the texture legacy-arm scope and tag the p7-magma remainder P13 2026-09-23 13:23:09 -04:00
swung0x48 e751b07baa [scripts] (P7): restore the executable bit the CI fix package dropped from junit_tally.py and stub_ctest.sh 2026-09-23 13:23:09 -04:00
swung0x48 cc37554a7d [MG_IntegrationTest] (P7): the cold-start entries hold the bring-up past the old 5 s bound, so they are red against the pre-fix client 2026-09-23 13:23:09 -04:00
swung0x48 877c531a0d [CI] (P7): tighten renamed CI titles 2026-09-23 13:23:09 -04:00
swung0x48 d7b9218618 [CI] (P7): short, readable job and step titles 2026-09-23 13:23:09 -04:00
swung0x48 e23e01e7ab [CI] (P7): one live Test and APK run per branch, and docs-only pushes skip the APK workflow 2026-09-23 13:23:09 -04:00
swung0x48 072ea1b832 [android-plugin, scripts] (P7): a passed APK replay whose device went offline before its logs were copied asks for the infrastructure retry 2026-09-23 13:23:09 -04:00
swung0x48 b13045918a [CI, scripts, tools] (P7): the pull control strips ctest -V's line prefix, retrace-split pins lavapipe, and the retrace summary reads role logs 2026-09-23 13:23:09 -04:00
swung0x48 2ab757792c [MG_Remote, MG_IntegrationTest] (P7): surface ops that may bring the spawned server's backend up wait a cold-start reply budget 2026-09-23 13:23:09 -04:00
swung0x48 cdf49a4624 [CI, tools] (P7): every retrace-split step gets the Magma repair knobs, and a verify retrace reads the pull library's mobilegl.log 2026-09-23 13:23:09 -04:00
swung0x48 15ce495363 [MG_IntegrationTest, CI] (P7): the dual-block census names the nine split-arm skips and both buffer lists require all twenty-one cases 2026-09-23 13:23:09 -04:00
swung0x48 76bfe8802a [MG_IntegrationTest] (P7): ObjectSubsystemControl reads the D3 refusal sentence and the four F1Wire decline cases are Magma-only 2026-09-23 13:23:09 -04:00
swung0x48 f1ce655b42 [MG_IntegrationTest, scripts] (P7): the Tcp TextureUploadShape entry holds the mobilegl-tcp lock, and the lane tools refuse any supervisor entry that does not 2026-09-23 13:23:09 -04:00
swung0x48 d1549cba57 [scripts] (P7): the TCP lane supervisor starts headless like every other server the harness reaches 2026-09-23 13:23:09 -04:00
swung0x48 eaef83b32a [MG_Remote, MG_Test] (P7): the server binds its deferred stream link through the Transport factory 2026-09-23 13:23:09 -04:00
swung0x48 14e1c8b9b1 docs: close p7w6 controls and F2 hardening status 2026-09-23 08:40:33 -04:00
swung0x48 e036b4e017 docs: record B4 landing and p7w6 matrix results 2026-09-23 06:07:36 -04:00
swung0x48 b19297d5e5 [MG_Backend, MG_Test] (P7): prove aggregate waits with a single-fence control 2026-09-23 03:53:37 -04:00
swung0x48 a4c1bc8b1e [MG_Backend, MG_Remote] (P7): draft B4 aggregate waits and staged extent reset 2026-09-23 03:53:37 -04:00
swung0x48 7ea7ce89cb [trace_replay] (P7): persist cancelled TCP attempts in results JSON 2026-09-23 02:12:40 -04:00
swung0x48 3ce5b9913e [docs, MG_Backend] (P7): record W6 device evidence and land M3 on review 2026-09-23 01:59:32 -04:00
swung0x48 9fd768d493 [MG_Backend, MG_IntegrationTest] (P7): bound wire descriptor sets within long frames 2026-09-23 01:35:20 -04:00
swung0x48 254bfb013e [docs] (P7): preserve the interrupted p7w5 matrix and p7w6 bsl evidence 2026-09-23 01:35:07 -04:00
swung0x48 f749d4c340 [trace_replay] (P7): pin the signed APK path used by a preinstalled device run 2026-09-23 01:08:18 -04:00
swung0x48 0e16ca272e [docs] (P7): HANDOFF-2026-09-22.md rewritten to the session's end state - nothing left to land, one matrix still running, the queue starts at p7w6 2026-09-22 23:15:28 -04:00
swung0x48 3374becde1 [docs] (P7): ID-P7-47/48 - V1 fix round 2 lands on the integrator's own review; the window-1b mid-run reading and the screen-off hypothesis are recorded for the next integrator 2026-09-22 23:11:14 -04:00
swung0x48 336174e36f [docs] (P7): the A-prime paragraph carries its where=entry counts
- the two counts were left as placeholders in the previous commit; they are 1 / 1 (DirectGLES / DirectVulkan), and the client summary of 4 / 3 is that one entry report plus the server read arm's 3 / 2, the divergence counter being process-wide
2026-09-22 23:05:20 -04:00
swung0x48 9814bc7426 [MG_IntegrationTest, MG_Remote] (P7): two citations stop misquoting - the old floor's unnamed count and a CMake line number
- split_log_paths.py's verify_split_arming docstring said the old count-only floor left ~117 entries unnamed; 117 is the lane's skip count, and the entries the floor left unnamed are the ones that did not arm - 155 on the last host run (1064 per-entry client logs, 909 armed), of which the 117 skips are a part
- CONTRACT-P7 §6 cited the monolith verify block by a CMakeLists line number ("约 :1715 起", with a note about an earlier wrong range); the guard `if (MOBILEGL_PIPE_VERIFY)` is the durable citation and now stands alone
2026-09-22 23:05:20 -04:00
swung0x48 07f4e3acef [MobileGL, MG_Backend, MG_Impl, docs] (P7): the FATAL=0 verify summary is written before Debug::Close, so the client half keeps its lines
- the summary line ("N divergence(s) survived MOBILEGL_PIPE_VERIFY_FATAL=0") came from a namespace-scope static's destructor, which runs from __run_exit_handlers AFTER MobileGL::Destroy has called MG_Util::Debug::Close(); Close nulls the role's sink, Log.cpp's WriteToFile re-runs InitFile() for the next line, and InitFile opens the role's file with "w" - so that one line was the only line the client half of every FATAL=0 run kept (121 B: the entry compare's reports, the arming line and the config dump all gone, on the split corrupt / read-corrupt entries and on the monolith arm's VerifyCorrupted. log alike)
- fixed in verify-only code, since Log.cpp is in the pull build and G1 admits no byte there: VerifyState gains FlushSummary(), which reports the count once (idempotent - a second call with nothing new, the destructor's own included, writes nothing); MGPipeVerifyFlushSummary() exports it from PipeFill.cpp, declared in PipeInputs.h beside the other verify-only entry points, and DestroyImpl calls it right before Close under #if MOBILEGL_PIPE_VERIFY
- the note's A' paragraph no longer says the entry compare is silent on the pack field - it reports it (the client summaries read 4 divergence(s) on DirectGLES / 3 on DirectVulkan); the red line can only come from the read arm because of where=read and the server-role file, and the paragraph now names the truncation and carries the numbers
- measured on the same entry, pipe-verify-split-read-corrupt-DirectGLES.client.log: 121 B / 1 line before, 7461 B / 85 lines after (DirectVulkan 121 B / 1 line -> 7386 B / 85 lines), with the where=entry report (1 each; the summary's 4 / 3 is that one plus the server's 3 / 2 read reports, the counter being process-wide) and the arming line present and the summary as the last line
2026-09-22 23:02:08 -04:00
swung0x48 ffc4fb7d77 [MG_Impl, MG_IntegrationTest, ci] (P7): the first perturbation block's falsifier is the DirectVulkan entry
- measured with one block removed at a time: without the re-applied block both server halves drop to 1 (DirectGLES 1 / DirectVulkan 1, both entries red); without the first block DirectVulkan drops to 1 (red) while DirectGLES stays at 2 (green), because the GLES backend reads the pack field twice after the neutral pack is installed (3 = the saved-pack read + 2 post-install) and DirectVulkan once (2 = 1 + 1)
- so the >= 2 requirement falsifies the re-applied block on both backends and the first block on DirectVulkan; the CI step runs both backends and needs both, which it already does
- the scenario, hook and CI comments now say exactly that, instead of "removing either block drops it to 1"
2026-09-22 23:02:08 -04:00
swung0x48 403b566949 [MG_Impl, MG_IntegrationTest, ci] (P7): the read-side control counts to two, one per perturbation block
- the server reads GetPixelStoreParameters twice inside the ReadPixels window - PipeApplier::read_pixels saves the application's pack through the accessor before it installs the neutral one, and the backend's ReadPixels reads the field after - and both reads sit inside ServerReadsInsideTheNeutralPackWindow; the hook's first perturbation is what turns the saved-pack read red (past the first compare the application's pack differs from the neutral oracle on its own), and only the perturbation re-applied after the window's overwrite can turn the post-install read red, because there the stored value is the neutral pack
- so a control that accepted ONE `where=read` line was satisfied by either block alone and falsified neither: the second block, which protects the lane's post-install backend reads, had no falsifier
- CorruptedFieldIsReportedOnTheServerRead now counts the `where=read` reports for the field in the server half since its mark and requires at least two; the CI read-side step does the same with grep -c; both block comments in the hook say which read each turns red
- measured on the host: both entries green, server half DirectGLES 3 / DirectVulkan 2; removing only the second block rebuilds green and turns both entries red with exactly one report per server half (1 / 1, ctest rc 8); the file was restored byte-identical and rebuilt
2026-09-22 23:02:08 -04:00
swung0x48 be764a56d7 [docs] (P7): ID-P7-46 - M2 round 2 lands on the integrator's own review, the memo handle-reuse ABA is closed 2026-09-22 23:02:07 -04:00
swung0x48 1ab62ece4c [ci] (P7): the parity exception list names M2 round 2's fourth Reclaim case
- ADrawAfterTheEarlyReclaimFollowsTheNewStoreNotTheMemoizedHandle rides the same split/spawn
  fixture as the three gauge cases and the tcp arm does not carry it (magma-m2.md §9.3); the
  integration tree keys this list by exact tail, so the fourth name goes in beside the three
2026-09-22 22:56:43 -04:00
swung0x48 f62703e933 [MG_Backend] (P7): a respecified wire store starts with no submission stamp
- M2 round 2 left this to the integrator: lastUseSubmitIndex is B3's field and exists only on
  the integration tree; without the reset the fresh store answers busy-by-submission until the
  old index retires (a needless staged copy, conservative not unsafe)
2026-09-22 22:51:20 -04:00
swung0x48 5ba34fcf42 [docs] (P7): magma-m2.md round 2 - the memo epoch fix and its red-once, the pipe tree's SHAs, the floor decision on its first reason alone, and the round's gates
- Section 0 cites the six commits by their pipe-tree SHAs (b81a8827 c569c916 0acee60a 9328bdc7 f0356dbe 5d8a2b99), with the branch SHAs alongside; every in-text cite follows.
- Section 2.1 keeps "the frame-serial floor is not a proof" on its first reason and drops the second, a stale citation of B3's fix; 6.3 is reworded to match.
- New section 9 records the review's must-fix (ID-P7-43): the handle-reuse ABA against UniformManager's memos, the destroy-epoch fix, the deterministic scenario and its red-once on both arms (the dead store's colour read back at base), the finding that the integration binary links the backend statically so a red-once variant has to be built into it as well (round 1's reds re-verified that way, unchanged), the three more knobs the spawn scrub keeps, the review's smaller items, the gates, and what is left for the integrator (the pipe-tree-only lastUseSubmitIndex reset in RespecifyWireBuffer).
2026-09-22 22:51:20 -04:00
swung0x48 6e18acaa5d [MG_Remote] (P7): the spawn launcher also passes MOBILEGL_IPC_SPIN_US, MOBILEGL_IPC_SERVER_AFFINITY and MOBILEGL_IPC_AUDIT through to the server child
- The environment scrub kept exactly one MOBILEGL_IPC_* knob (WIRE_DEFERRED_MB, M2); these three are also read by the server and were silently defaulting in the spawned child for the same reason: SPIN_US is the doorbell spin budget both roles read, SERVER_AFFINITY the apply thread's CPU mask, AUDIT the 0xDD fill of retired stage bytes. Each is a tuning or audit value; none names an endpoint, a role, a path or a segment size, and ServerMain's catch (b) keeps verifying the four it verifies (TRANSPORT, SERVER_PATH, RING_MB, STAGE_MB).
- Everything else under the prefix still goes (the fable review of M2, item on ServerSpawn.cpp).
2026-09-22 22:51:20 -04:00
swung0x48 959ebc9e76 [MG_IntegrationTest] (P7): MagmaWireReclaimScenario draws through a handle the early reclaim freed and re-minted, and reds on the dead store's bytes; the count-ceiling case refuses an absent gauge
- ADrawAfterTheEarlyReclaimFollowsTheNewStoreNotTheMemoizedHandle is the round-2 red-once for ID-P7-43, deterministic at every step: a draw from a 256-byte UBO store memoizes its handle; glBufferData(0) orphans it without minting; glClientWaitSync with the flush bit and a zero timeout submits the batch without waiting (the review's flush-without-wait); a glBufferSubData into the busy vertex store records a staged copy so the blocking glClientWaitSync that follows cannot take the frame-boundary drain that clears the memos; the vertex store's own glBufferData(0) is the park whose sweep destroys the dead store; a 64 KiB glBufferData through the same GL buffer mints through the freed handle value (glibc tcache, LIFO) into memory that cannot alias the 256-byte hole a 1 MiB pin store bounds; the second draw resolves the same (handle, range). The pixel must be the new store's colour; at base it is the dead store's.
- Red-once (rule J), the epoch guards removed from UniformManager.cpp and the variant built into MobileGL, MobileGLServer AND MobileGLIntegrationTest: Split and Spawn both read rgba(200,60,30,255) - the dead store's colour - against the wanted rgba(60,200,90,255); green with the guards. The integration binary links the backend statically (MobileGL_s), so a variant built into the two shared targets alone never reaches the inproc arm - the first run of this red-once read Split green for that reason, and a re-run of round 1's base variant into all three targets reproduces round 1's reds on both arms unchanged (1025 vs 4; 49,287,168 vs 9,437,184 and 49 vs 11; 3001 vs 1027).
- ManySmallRespecifyAndDrawRoundsStayUnderTheStoreCountCeiling asserts the wlivepk gauge is a reading before bounding it, as the other two cases already did: an absent gauge read -1 and passed the one-sided bound as a leak that never happened.
- Registered on the Split and Spawn Reclaim entries beside the three; spawn_lane_parity.py's MAGMA_NOT_ON_TCP names the fixture by prefix, so nothing there moves.
2026-09-22 22:51:20 -04:00
swung0x48 4d20746831 [MG_Backend] (P7): the descriptor memos fold in a wire-store destroy epoch, so a VkBuffer handle the mid-frame reclaim freed and the next mint handed back cannot revive a set baked to the dead store
- M2's defer path destroys a wire store MID-FRAME (the serial-0 park, the sweep's prefix walk, the watermark); before it a wire store only died at the boundary that also clears UniformManager's memos (BeginFrame from Present or a drain). Those memos are keyed on the VkBuffer HANDLE - m_descriptorReuseMemo hashes the VkDescriptorBufferInfo words, FastRebindMemo compares uboBuffer - and on the wire arm the direct-bind UBO and the SSBO descriptor name the store's own handle. A store destroyed at a park, its handle value re-minted by the next Create (a heap pointer under lavapipe, LIFO out of glibc's tcache), and a draw resolving the same (handle, range) reused a set whose descriptor still pointed at the dead store's memory: silent wrong bytes, no Fatal (ID-P7-43, the fable review's must-fix).
- VkBufferManager gains m_wireStoreDestroyEpoch, bumped on every path that destroys a wire store's VkBuffer (DeferWireRelease's immediate destroy, SweepDeferredWireReleases once per sweep that freed anything, DestroyAllDeferredWireReleases, Shutdown's record teardown) and never reset, with GetWireStoreDestroyEpoch(). UniformManager mixes it into the descriptor-reuse signature and records it in FastRebindMemo, refusing the fast rebind when it moved; all under MOBILEGL_BUILD_DISAGGREGATED. Deliberately not m_sliceEpochCounter, which every WriteWireBuffer bumps and would defeat the memo on every glBufferSubData. On the wire arm only the signature is reached (SetupWireDraw passes no sampler hint), the FastRebindMemo field is the defensive half.
- Two comments from the same review: SweepDeferredWireReleases no longer claims GetSyncPointSubmitIndex() never decreases - it steps back by one across an abandoned recording (RecreateSwapchain, the minimized Present), and the prefix walk is conservative across that, not wrong; and the header's "floor is not a proof" keeps its decision on its first reason alone, the second was a stale citation of a fix that is on the integration tree.
- Section 9 partition exception for UniformManager.{h,cpp} granted by the integrator for this fix.
2026-09-22 22:51:20 -04:00
swung0x48 efa19cb16f [docs] (P7): ID-P7-45 - X2 lands: the InitialCapsStartup flake was the fixture pairing loopback sockets by accept order 2026-09-22 22:20:50 -04:00
swung0x48 9a4752adba [docs] (P7): X2 - note for the InitialCapsStartup flake: TCP accept order, not the caps bring-up
- docs/Disaggregated/notes/p7/x2-caps-flake.md: the failure text (Fatal{ProtocolCorruption,
  "StreamLink"} invalid data frame header, SIGABRT), the mechanism (ConnectTo/AcceptPair pairs
  two loopback TCP connections by accept order and the kernel reorders them under load: 5 of
  3000 in a raw-socket probe), the TEST verdict against the product's DataBind rendezvous, the
  before/after loop counts (2/500, 5/500, 3/500 -> 0/500 x3 under the same load) and the gate
  numbers (unit 2435/2435, census 79, ratchet 173, G1 .text 0xa52203 and +0/-0 symbols)
- open: AcceptPair/ConnectTo over tcp:// are still callable and still pair by order; refusing
  them by name is a product change with its own red-once
2026-09-22 22:20:24 -04:00
swung0x48 e730a91206 [MG_Test] (P7): X2 - SocketTransportTest's TCP case uses the DataBind rendezvous too
- TcpCarriesControlAndIndependentDataWithKeepalive had the same assumption as the
  InitialCapsStartup fixture: ConnectTo/AcceptPair over loopback TCP, paired by accept order;
  when the kernel reordered the two connects the bytes sent on the client's data socket arrived
  on the server's control socket and recv(serverData, MSG_WAITALL) hung until the 60 s kill -
  3 of 500 runs under the same load, 0 of 500 afterwards
- the case now opens the control connection, accepts it, then opens the data connection with a
  DataBind nonce and decodes it on the accepted side, exactly the shape the product uses on TCP;
  it keeps its NODELAY, keepalive and ShareFd-unsupported checks on both control transports and
  its independent-data check on the bound data sockets, and asserts that a control-only TCP
  transport owns no data descriptor (TakeDataFd == -1)
2026-09-22 22:20:24 -04:00
swung0x48 f91d0f2c4b [MG_Test] (P7): X2 - InitialCapsStartup pairs its TCP sockets by DataBind nonce, not by accept order
- the fixture rendezvoused with ConnectTo/AcceptPair over loopback TCP, which pairs the two
  connections by the order the listener accepts them; under host load the kernel delivers two
  back-to-back loopback connects to the accept queue out of connect order (measured 5 of 3000
  with 20 CPU hogs), the peer's control socket became the client's data socket, the first control
  frame hit the StreamLink reader and the process died of Fatal{ProtocolCorruption, "StreamLink"}
  invalid data frame header - 2 of 200 and 2 of 500 runs of ANullInitialSnapshot..., 5 of 500 of
  DelayedTcpSnapshot..., every one in 3 of the last 4 integration gates on the loaded host
- the flake is in the TEST: PH-7 (4) already took by-order pairing out of the product on TCP
  ("never paired by arrival order"); the fixture now does what ServerMain's ListenerSource does -
  ConnectControl is accepted before ConnectDataConnection is opened, and the data connection
  names itself with a DataBind nonce that the peer decodes and matches in constant time
- 0 of 500 for both cases under the same load afterwards; AHealthyPeer... shares the fixture and
  is covered by the same change; no product bytes touched
2026-09-22 22:20:24 -04:00
swung0x48 bf8e605fd0 [docs] (P7): ID-P7-44 - the integrator's own reviews land the F fix round and the B3 r3 / B2 r4 rounds, with two small debts recorded 2026-09-22 22:16:07 -04:00
swung0x48 137c911a94 [docs] (P7): ID-P7-43 - M2 lands with a must-fix routed to M2 round 2, and HANDOFF-2026-09-22.md lists every mechanical step left for the next integrator 2026-09-22 22:08:23 -04:00
swung0x48 63347c1682 [docs] (P7): ID-P7-42 - M2 and the F fix round pass their gates, B3 round 3 and B2 round 4 join the tree, and the verify counts are the integration tree's 2026-09-22 22:05:21 -04:00
swung0x48 4b96f55d32 [docs, MG_IntegrationTest] (P7): magma-b2.md carries review round 4 - the transport gate loses a conjunct nothing reaches, and the debt lines cite by symbol
- review round 4 nit: the second gate of the flipped multisample resolve case read `!SplitLane::IsSplitLane() && !PeekSplitRuntime().transportResolved`; the marker term is dead - with the marker set and no transport resolved, the fixture's SetUp already skipped through SplitLane::SkipReasonForSplitOnlyAssertions() -> SplitRuntimeSkipReason(), so Ready() is false and the gate is never reached; in every reachable state it equals `!PeekSplitRuntime().transportResolved`, which is what it now says (the SplitLane.h include goes with it), and the comment no longer describes an unreachable branch; behaviour unchanged - monolith entries still Skipped with the same text, the Full entries still Passed
- review round 4 nit: magma-b2.md §6 items 1, 2, 3 cited WireFramebuffer.inc by line numbers that round 3 had already moved; they now name the arm inside BlitWireFramebuffers's `blit` lambda and the MGLOG_E_ONCE / RecordTextureCopyError text, and the monolith cross-reference names VulkanRenderer::BlitFramebuffer's vkCmdResolveImage + vkCmdBlitImage instead of pipe line numbers
- §2.11 records the round: the should-fix (window primed magenta/0.625, the one-shot wrong-surface demonstration and its readings), the two nits, and a round 4 line under §7 with this tree's readings
- test and docs only; no backend source touched (G1 unaffected)
2026-09-22 22:03:23 -04:00
swung0x48 e385ea96c8 [MG_IntegrationTest] (P7): the window leg of the flipped multisample resolve case primes the default framebuffer with values no other surface in the test holds
- review round 4 should-fix: round 3 primed the window blue at depth 0.5, the same colour and depth the user framebuffer `resolved.fbo` was primed with and still held after its own declined blit, so a glBindFramebuffer(GL_FRAMEBUFFER, 0) that landed on the user framebuffer, or a default-framebuffer readback that resolved the wrong surface, read the priming it was meant to reject and the leg passed vacuously
- the window is now primed magenta at depth 0.625 - the user framebuffer is blue at 0.5, the source red/green at 0.25/0.75 - and the three probes expect exactly that
- shown once and restored: with the probes pointed at `resolved.fbo` instead of 0, all three arms (`DirectVulkan.{Split,Spawn,Tcp}.MsFlip.`) fail on both aspects at (1, 1): colour reads {0,0,255,255} against {255,0,255,255} ("must not have written its colour at (1, 1)"), depth reads 0.5 against 0.625 ("must have left its depth alone at (1, 1)"); under round 3's priming the same probes on that surface were green
- test only; no backend source touched (G1 unaffected)
2026-09-22 22:03:23 -04:00
swung0x48 c8fc4fb712 [docs] (P7): magma-b3 - fix round 3: the audit's guard-aware walk, the tcp-naming server-log fatal, the EVIDENCE guard, and the round's gate
- §0: rows 28-31 for this round
- §1.5: the two walk holes the third review found (a log under `#if 0` accepted, an indented
  `#else` walked through), the fail-open-for-one-guard rule, and the two new fixtures' readings
  on the round-2 audit and on this one; 16/16, tree 53 / 53 / 0; the probe shapes that are not
  fixtures
- §4.1: the merged server-log fatal names MOBILEGL_IPC_CONTROL; demo 1 re-run on the new runner
- §4.5: the EVIDENCE empty-pattern guard; the stub's executable bit restored
- §7.2: the B4 sentence gives both numberings (package tree :13404 / :13471, landed tree
  :13533 / :13600); two debts recorded - the tcp pid-arm-proof check still runs before the
  marker check, and LOG_WINDOW counts the blank and `#if` lines it skips
- §5: the round's gate on the code head; no product file, so G1 is unaffected; lanes not re-run
2026-09-22 22:03:23 -04:00
swung0x48 3b8bf9ec25 [scripts] (P7): the pull-library control refuses an empty grep -F pattern in EVIDENCE, and the stub ctest is executable again
- every line of EVIDENCE is a grep -F pattern and an empty line matches everything: a trailing
  newline or a blank line between the two sentences would turn hole 2 into "any red will do".
  The smoke's "red without the transport-resolution message" case catches that (22/23), but
  only when somebody runs it; a `case` after the assignment now refuses a blank line or a
  leading/trailing newline with ::error:: and rc 1, before the selection is counted or the
  library is swapped. The red-check's perturbation line is untouched; smoke 23/23
- scripts/ci/testdata/stub_ctest.sh went 100755 -> 100644 in the round-2 commit that added
  its retrace-evidence-nolog mode (a Windows-mount artefact); restored with
  `git update-index --chmod=+x`. The smoke copies it and chmods the copy, so nothing ran
  differently
2026-09-22 22:03:23 -04:00
swung0x48 10220b564e [tools] (P7): the merged server-log fatal names MOBILEGL_IPC_CONTROL, so the tcp arm's red still says tcp
- the round-2 merge folded "TCP session has no forwarded server log" into the one server-log
  check for every transport, and its message named only MOBILEGL_TRANSPORT - which is `spawn`
  for the tcp arm too (tcp = spawn + MOBILEGL_IPC_CONTROL=tcp://), so the tcp arm's missing
  forwarded log read as a plain spawn red
- the message carries MOBILEGL_IPC_CONTROL='...' after the transport (empty for inproc and
  plain spawn); message only, no check moved
2026-09-22 22:03:23 -04:00
swung0x48 bd849b2821 [tools] (P7): the audit's block walk is fail-open only for the disaggregated guard, and an indented #else stops it
- stepping over `#endif` accepted a log that is compiled out: `#if 0` / `MGLOG_W(...)` /
  `#endif` / `Count(...)` read as logged, and so did a log under `#ifdef
  MOBILEGL_VERBOSE_DECLINES` - a Count that counts every time and logs in no shipped build
- the walk now counts the `#endif`s it crosses; a same-indent log met with one outstanding
  counts only if every opener enclosing it, reached by continuing upward, is the whole-line
  `#if MOBILEGL_BUILD_DISAGGREGATED` (`&& X` is not it). `#if`/`#ifdef`/`#ifndef` with no
  `#endif` outstanding is the Count's own guard and is still stepped over, so the column-0
  `#if/#endif` fixture stays green; the comment says the walk is fail-open for that one
  guard and nothing else
- `#else`/`#elif` stopped the walk only at column 0, as a line of lesser indentation; one at
  the Count's own indentation walked straight through to the log above it. PP_STOP now
  returns False at any indentation, before the indent test
- two red fixtures, both green on the round-2 audit: "log only under a #if 0 above the
  Count" and "log only on the other side of an INDENTED #else at the Count's indentation";
  16/16, the tree stays 53 / 53 / 0
2026-09-22 22:03:23 -04:00
swung0x48 9cea514010 [docs] (P7): ph-f.md §6/§7 record the F fix round - the fingerprint refusal is one-directional and an un-redeployed phone shows as "no Welcome", the unix pairing window and the two pre-auth exposures are §12 debts, the token minimum is TCP-only, and the token examples are 16+ bytes
- ph-f.md: §0 rows F-a..F-d with package-tree SHAs; §2.2 states the asymmetry (new server + old client -> Refuse{WireFingerprint}; old server + new client -> the old AcceptPair closes both after 2000 ms and the client sees `no Welcome ... (rc=6)`, no Refuse frame); §6 the four fix commits with their red-once lines; §6.4 the unix arrival-order pairing (250 ms / 2000 ms) as a local-peer debt with the SELinux assumption named, and the two pre-auth exposures; §7 the gates on 2b2a0ae0
- CONTRACT-P7 §12: the two pre-auth exposures (fork before Hello with a 10 s wait; ValidatePeerHandshake before AuthenticatePeerToken sends the fingerprint and build stamp to an unauthenticated peer), the unix pairing debt, and the fingerprint asymmetry, for PH-7 (5)
- PLAN-PH-P34B-P7: PH-7 (3) says the minimum is checked at the TCP listen; PH-8 says "ignored, server-sized" rather than "clamped"; the no-token listen row's remaining-work cell no longer claims the smoke is unreferenced
- device-window-1-runbook §2 step 3: the un-redeployed-phone symptom, and where to look first
- ROADMAP §6 and peer-loss-probe: `MOBILEGL_IPC_TOKEN=devtoken` (8 bytes, which now exits 72) becomes a 25-byte token in every example
2026-09-22 21:55:39 -04:00
swung0x48 8c8fc31952 [MG_Remote, MG_Test, ci] (P7): the token compare runs a fixed 64-byte width, an ignored window ask is logged once, a nonce that could not be minted is refused by name, and a garbage first frame while busy is MalformedHello rather than Busy
- AuthToken.h: ConstantTimeTokenMatch loops kAuthTokenCompareBytes (64) iterations with both operands read as zero past their ends - its trip count no longer tells a peer the configured token's length; a token longer than 64 is still compared to its last byte (SocketTransportTest pins the 70-byte case)
- ServerSession.cpp: an ask in Hello.linkTerms above the granted terms logs one MGLOG_W naming both sides ("the ask is ignored and the session is server-sized"); the production client asks for nothing, so a session that was never mis-configured never logs it; SessionHandshakeTest's 64 GiB case requires the line (red before: absent)
- ServerSession.cpp: a MintNonce failure sends Refuse{LinkTerms} "stream data plane unavailable: the server could not mint a data nonce" before returning the fault it is (still exit 67, still counted) instead of a bare close
- ServerMain.cpp: a first frame with the wrong magic or over 1 MiB while a session is live is Refuse{MalformedHello} "first frame is not a control frame", drained before the close so the refusal is not RST away; Busy stays the answer for a well-formed frame that is not a DataBind
- tcp_supervisor_smoke.py: garbage_first_frame_while_busy (magic XXXX on a live session -> code 8; red before: code 7)
- fatal census unchanged: 79 abort sites, 3 refusal words (the new refusals go through the enum, not a literal)
2026-09-22 21:54:46 -04:00
swung0x48 c69d8dfdc8 [MG_Remote, MG_Test, ci] (P7): sha256(protocol.fbs) is pinned to MOBILEGL_PROTOCOL_CONTROL_REVISION, so a control-schema edit without a bump is red in the unit lane and in flatc-check instead of a silently absent field
- the revision is a hand-bumped integer mixed into wireFingerprint and nothing tied it to protocol.fbs's content; scripts/ci/protocol_revision_pin.py holds {revision -> sha256(protocol.fbs)} in protocol_revision_pins.json and refuses a schema whose digest is not the current revision's, or a revision with no row
- registered as ProtocolSchema.RevisionPinsDigest (label unit, MG_Test/Wire/CMakeLists.txt, --self-test proves both perturbations red before it says green) and as a step of CI's flatc-check job, the one that regenerates the header on every schema edit
- --write records a bump's row and refuses to re-pin an existing revision to a different digest, which is what a schema edit without a bump looks like
- mg_protocol_base.h says where the pin lives beside the revision it pins
- red-once: protocol.fbs touched with one trailing comment, revision left at 1 -> "protocol.fbs has sha256 a7b5c7fb... but MOBILEGL_PROTOCOL_CONTROL_REVISION is still 1, which is pinned to 7fc859f7..." rc=1; restored, green
2026-09-22 21:54:46 -04:00
swung0x48 0293aebb17 [MG_Remote, ci] (P7): a bound session closes its data-connection source, so a later DataBind is refused by name instead of queued into a child that stopped reading, and the single-session listener closes with it
- ServerMain.cpp: RunSession takes the descriptor its data-connection source reads from (the child's hand-off end under --serve, the listener itself without it) and closes it the moment Accept returns; the source object goes with it
- the supervisor's hand-off send is MSG_DONTWAIT (FdPassing::SendFd dontWait), so a peer that stopped reading answers at once; "peer gone" on a datagram pair (ECONNREFUSED, ENOTCONN) is an answer the caller names, not a logged fault
- before: every DataBind after the bind sat in the child's receive queue with no reply, its TCP connection held open until the session ended, and past the queue limit (net.unix.max_dgram_qlen 10 on the phone; ~288 datagrams of sender buffer on this host) the supervisor blocked in sendmsg for the rest of the session - no accepts, no Busy, no reaping
- before, without --serve: the listener stayed open after the data connection bound, so a second client or a retry completed a TCP handshake nobody would accept and waited out its 20 s connect budget instead of the ECONNREFUSED it got before PH-7 (4)
- tcp_supervisor_smoke.py: data_bind_after_bound (12 extra DataBinds on a bound session each get Refuse{Authentication} "data connection could not be handed to the live session", and a fresh control connection is still Busy) and single_session_listener (a connect after `ready` is ECONNREFUSED); connect_control retries ECONNREFUSED at startup for pair() and held_session() alike
- red-once at da313010: smoke TimeoutError on the first extra's receive; probe with qlen+2 fire-and-forget DataBinds: 288 queued silently, then the supervisor stopped accepting (fresh control: connect timed out) while still alive; single-session connect after ready succeeded
2026-09-22 21:54:46 -04:00
swung0x48 9ae1e6c04e [docs] (P7): ID-P7-41 - M2 is on the integration tree with its second leak named, three reviews land with fixes, and window 1b's instant failures are the pairing window over the LAN 2026-09-22 21:48:27 -04:00
swung0x48 7d1315271a [docs] (P7): magma-m2.md names the one server-side-knob parity list its three entries joined at merge 2026-09-22 21:46:08 -04:00
swung0x48 5d8a2b99fd [docs] (P7): magma-m2.md records M2 - the wire arm's orphaned VkBuffers are bounded (25,935 -> ~1,050 peak, bsl spawn server 825 -> 546 MiB), and the host maps M1 attributed to them are lavapipe descriptor sets
Before/after on host with M1's sampler and probe for bsl-esc-menu-854 and iris-bsl-in-world on
monolith / inproc / spawn (pixels identical on every arm), the mechanism with citations, the
knob's semantics, the red-once table, the gates, and what remains: the wire arm's descriptor-set
growth between BeginFrame rewinds (27,377 live sets == 28,134 maps; now bounded only because
M2's forced syncs let the drain rewind the cursors), CollectGarbage/pruneWire never reached on
the wire arm (B4), the frame-serial proof deferred until B3's floor fix, the tcp arm's gauge
reach, the ServerSpawn.cpp touch and the knob for the integrator to ratify, and the Redmi
acceptance run.
2026-09-22 21:46:08 -04:00
swung0x48 f0356dbe61 [MG_Backend, MG_IntegrationTest] (P7): the wire watermark's sync point also fires above 1024 parked stores, so small orphans that never reach the byte budget are bounded by count
Measured on bsl-esc-menu-854 (spawn, lavapipe) with the byte watermark alone: the dead VkBuffers
at the end fell from 25,935 to 32, but one stretch with no submission still parked 12,498 stores
in 39.5 MB - under 64 MiB, so nothing forced a sync. A fixed ceiling
(VkBufferManager::kWireDeferredCountCeiling = 1024, not a knob) now rides the same
WaitForSubmitIndex call; MOBILEGL_IPC_WIRE_DEFERRED_MB=0 still turns both triggers off.
MagmaWireReclaimScenario gains the case: 3000 respecify-and-draw rounds of a 256-byte store in one
frame (750 KB, far below the lane's 8 MiB). Red before this commit on inproc and spawn (live peak
3001 vs a bound of 1027), green after (1025, two forced syncs).
2026-09-22 21:46:08 -04:00
swung0x48 9328bdc735 [MG_Backend, MG_Util, MG_Test, MG_IntegrationTest, scripts] (P7): integration-magma-split and -spawn carry MagmaWireReclaimScenario, which reads the server's wbuf[] gauges and reds on a wire arm that keeps orphaned buffer stores until the frame ends
Rule J's red-once for M2. The server role's VkBufferManager publishes four PipeStats gauges
(push builds only, so the pull image is untouched): wire buffer records alive, and the run
maxima of live wire VkBuffers and of parked bytes, plus the watermark's forced-sync count -
maxima because the defect lived inside one frame and a swap-time sample reads after the frame
boundary's own sweep. Two cases, each inside a single frame: 1024 undrawn respecifies of one
4 KiB buffer (live peak <= 4 x wire buffers), and 48 respecify-and-draw rounds of a 1 MiB store
under the lane's MOBILEGL_IPC_WIRE_DEFERRED_MB=8 (parked peak <= budget + one store, every strip
its own colour). Red at base (ec46a550's defer path, counted): 1025 vs 4 live, 49,287,168 vs
9,437,184 bytes and 49 vs 11 live, on both arms; green after: 1 live, 9,437,184 bytes / 9 live
with 5 forced syncs. The knob's negative control (=0) reds the second case with the same
49,287,168 bytes. Inproc and spawn only: the tcp server is the lane-wide fixture supervisor,
whose environment and log the entry cannot reach - named in spawn_lane_parity.py as
MAGMA_NOT_ON_TCP (the ID-P7-14 shape), and PipeStatsTest pins the four short names.
2026-09-22 21:46:08 -04:00
swung0x48 0acee60a3b [MG_Remote] (P7): the spawn launcher passes MOBILEGL_IPC_WIRE_DEFERRED_MB through to the server child, the one MOBILEGL_IPC_* knob its environment scrub keeps
BuildChildEnv drops every MOBILEGL_IPC_* entry, so the server-owned watermark meant something
under inproc and silently nothing under spawn - measured: the spawn server parked 47 MiB against
a lane budget of 8 with zero forced syncs, because it had parsed the default out of an
environment that no longer carried the variable. The knob names no endpoint, role or segment
size, so neither anti-recursion catch (Dial=No, ServerMain's scrub check of TRANSPORT /
SERVER_PATH / RING_MB / STAGE_MB) is weakened; it is listed by name, not by widening the prefix.
2026-09-22 21:44:06 -04:00
swung0x48 c569c916c0 [MG_Backend] (P7): when the orphaned wire buffer stores still parked exceed MOBILEGL_IPC_WIRE_DEFERRED_MB (64 by default), the server takes the host-access sync point mid-frame and reclaims them all
The defer-path sweep frees an orphan once its last submission has retired, but inside one long
frame a store that was drawn from is named by recorded, unsubmitted work, and nothing submits it
until the frame ends. The knob is read once through MG_Config::Ipc like its siblings; above it
the park calls WaitForSubmitIndex(GetSyncPointSubmitIndex(), flush) - the call
WaitForWireBufferHostAccess makes - and sweeps again, which empties the list because no entry can
be tagged past the sync point just waited out. A sync point that cannot complete dies through
the existing Fatal{ResourceUnavailable, "buffer-write-sync"} site, now one helper shared with the
host write (fatal_census: 79 abort sites, unchanged). 0 is the negative control.
2026-09-22 21:44:06 -04:00
swung0x48 b81a8827b5 [MG_Backend] (P7): a wire buffer store that glBufferData orphans is destroyed on the defer path as soon as no GPU command can still name it, instead of waiting for a frame boundary
RespecifyWireBuffer (every wire glBufferData) and DestroyWireBuffer parked the old VkBuffer in
the per-frame-slot buckets, which only BeginFrame and the idle drain empty. A snapshot-exiting
pbuffer replay delivers one present per run (ID-P7-32), so on bsl-esc-menu-854 the split server
held 25,923 dead stores. The wire arm now keeps its own flat list: a store whose last-use serial
is 0 (never named by a GPU command) is destroyed at the park; any other is tagged with the
renderer's sync-point submit index and destroyed by the sweep every park runs, once that index
has completed or the renderer is idle. The frame-serial floor is deliberately not a proof here
(it cannot move mid-frame, and on this base it is the floor B3 found unsound). The transient
arena's releases and the monolith OnRespecify path are unchanged.
2026-09-22 21:44:06 -04:00
swung0x48 41b392298b [docs] (P7): ID-P7-40 - window 1b's TCP matrix runs on the user's route, and the B3, V1 and B2 fix rounds are on the integration tree 2026-09-22 21:38:13 -04:00
swung0x48 11edc8e3ac [docs, MG_Backend] (P7): magma-b2.md carries review round 3 - the transport gate, the default draw framebuffer in the pre-pass, the shape-only scope of "a declined call writes nothing", seven knob tails, and three new debt lines
- §2.10 (new): the skip gate is the process's resolved transport, with the Skipped/Passed lines as measured (monolith + DirectGLES Skipped, Split/Spawn/Tcp.Full Passed); why VerifySplit. cannot be measured on this branch (a -DMOBILEGL_PIPE_VERIFY=ON build of this tree aborts every split entry's bring-up in MGPipeApplyResourceRespecify on the apply thread - the verify+split debt the verify-split package relaxes) and its measurement on a throwaway detached worktree of the landed head plus commit a (Passed); the default draw framebuffer joining the pre-pass, with the red-once (round-2 guard restored -> Split/Spawn MsFlip. read the source's red/green in the window); the two wording fixes.
- §2.9 items 1 and 3 and §6 items 1 and 5: the entries that now RUN the case (Full. x3, VerifySplit. on the landed tree), the corrected default-draw sentence, and "a declined call writes nothing" scoped to SHAPE declines - the capability decline in ResolveWireDepthStencil is still per aspect, rule I (a).
- VulkanRenderer.h: the DeclineWireDepthStencilResolveShape comment says the same (comment only).
- §6 item 4 and §7: seven server-env-knob tails on the landed tree, naming .StaleSerial. (this branch predates B3 and measures six).
- §6 items 8-10 (recorded, not fixed): a legal same-size MSAA-FBO -> default-framebuffer depth/stencil blit is a session Fatal (multisample-depth-resolve-region); a scissor that empties the destination returns silently even for shapes GL still errors on; MS->MS blits decide per aspect, so a COLOR|DEPTH call can copy colour and decline depth.
- §7 is one measured row for this tree (p7/magma-b2@1bf48cfb + this commit's comment, remeasured on the final head) and names the landed tree as the integrator's gate, without numbers or SHAs.
2026-09-22 21:36:08 -04:00
swung0x48 24da1e0f6d [MG_Backend, MG_IntegrationTest] (P7): the default draw framebuffer is asked the multisample depth/stencil shape question before any aspect runs, so a scaled COLOR|DEPTH resolve onto the window writes nothing
- BlitWireFramebuffers' round-2 pre-pass left a default DRAW framebuffer to the per-aspect lambdas on the belief that a default draw behind a multisample read was ResolveWireDepthStencil's region Fatal whatever the shape. It is not: that arm calls DeclineWireDepthStencilResolveShape BEFORE its destination.isDefault region check, so a scaled (or X-mirrored) COLOR|DEPTH blit out of a multisample framebuffer onto the window ran the colour lambda first (BlitWireColorToDefault wrote the swapchain image) and then recorded INVALID_OPERATION on its depth - the partial effect magma-b2.md §2.9 item 3 declared closed.
- The guard drops !draw->IsDefault, and the pre-pass resolves both sides as READS (isWriteTarget=false): only their format and samples are consulted, and DefaultFramebufferReadIndex is const, so nothing re-points the default framebuffer at the acquired image before a write.
- The flipped-resolve case gains a leg: the window is primed blue at depth 0.5, the multisample source (red below, green above) is blown up over the whole of it with COLOR|DEPTH, the call must be INVALID_OPERATION, and three probes (corner, centre, far corner) must read the priming on both aspects through glReadPixels - the only way the harness's pbuffer default framebuffer can be looked at.
- Red-once (split + spawn, restored): the round-2 guard (&& !draw->IsDefault) put back -> DirectVulkan.{Split,Spawn}.MsFlip. fail "the declined COLOR|DEPTH scale onto the window must not have written its colour at (1, 1)" reading {255,0,0,255} (the source's bottom band), and {0,255,0,255} at (64, 48) and (126, 94), for {0,0,255,255}; the INVALID_OPERATION assertion still held, which is the partial effect exactly.
- Verified: the 107 multisample blit/resolve and depth/stencil readback entries on the three wire arms (MsFlip/MsFlip1/MsBlit/MsResolve0/1, Full x3 for both readback scenarios) and the monolith DirectVulkan.DepthStencilReadback* entries pass, pre-existing skips unchanged.
2026-09-22 21:36:08 -04:00
swung0x48 6f1c6d26ee [MG_IntegrationTest] (P7): the flipped multisample resolve case skips on the transport the process resolved, not on the lane marker, so the whole-binary census entries run it again
- MGITEST_SPLIT_LANE marks the CURATED split lanes, not the transport: DirectVulkan.{Split,Spawn,Tcp}.Full. (and the VerifySplit. lane where it is registered) run inproc/spawn/tcp WITHOUT it on purpose, and the round-2 marker gate skipped the case there with a message about a monolith arm that was not running - before round 2 it ran and passed on them.
- The gate is now the process's resolved transport (PeekSplitRuntime().transportResolved, i.e. MG_Config::Transport != Monolith), which is the very fork VulkanRenderer::BlitFramebuffer takes to reach BlitWireFramebuffers; the marker on its own still keeps a curated lane running the case, and its fixture's arming assertion says whether that lane got the transport it asked for.
- Verified on build-split: DirectVulkan.DepthStencilReadbackMatrixScenario.AFlipped… still Skipped ("the monolith DirectVulkan transport … refuses a flipped depth blit …"); DirectGLES.{,ForcedDepthStencilEmulation.}…AFlipped… still Skipped (Espryt reason); DirectVulkan.{Split,Spawn,Tcp}.Full.…AFlipped… Passed (all three were Skipped before); DirectVulkan.{Split,Spawn}.{MsFlip,MsFlip1}. and Tcp.MsFlip. Passed.
2026-09-22 21:36:08 -04:00
swung0x48 0bc59ab239 [MG_IntegrationTest] (P7): the unarmed-expected file stops misquoting tier 3s count
- the first class said its 22 names were "the same 13 entries per backend that tier 3 measured as reds", which is wrong twice: 13 was the whole count, not a per-backend one, and those 13 (AdvertisedLimitsScenario x8, SpirvShaderBinaryScenario x4, ProgramPipelineScenario's reserved name) are spread across this class and the next
- the sentence now says what tier 3 actually measured and why it is the same property seen from the other side
2026-09-22 21:36:08 -04:00
swung0x48 16b1d031a9 [docs] (P7): the note's lane tables carry the V1 fix round's own numbers
- integration-verify goes 1136 to 1138 (273 skip): the new CorruptedFieldIsReportedOnTheServerRead case is discovered by the ambient Verify. registration too, where it skips for want of the CORRUPT knob - the same shape CorruptedFieldIsReported already has there
- integration-verify-split is 1072/1072 with 117 skip, re-measured on the final head
- 0 gains a row for the fix round, one line per finding, pointing at the section each landed in
2026-09-22 21:36:08 -04:00
swung0x48 c0f7aa9bd6 [docs, MG_Impl] (P7): the pack blind spot is one value, not the field
- 2.2 and the hook's own comment both said the comparator can no longer see a client that pushed a WRONG pack state on the server's ReadPixels. Too broad: PipeApplier.cpp reads savedPack THROUGH THE ACCESSOR, and therefore through the hook, BEFORE it installs the neutral value, so the pushed pack is compared against the live one right there
- a correct push is equal and returns; a wrong push that is not exactly the neutral pack fails both compares and is still Fatal; only a wrong push that equals {SwapBytes 0, LSBFirst 0, RowLength 0, ImageHeight 0, Skip* 0, Alignment 1} is invisible, and only while the application's own pack differs from it
- notes/p7/verify-split.md 2.2 and 6 say so, and the comment above ServerReadsInsideTheNeutralPackWindow says so
- the measured evidence is the read-side control's own server half: 2 to 3 where=read reports per readback, which is the savedPack read plus the backend read
2026-09-22 21:36:08 -04:00
swung0x48 3947a344b4 [docs] (P7): CONTRACT-P7 6 and 8 stop contradicting themselves about verify x split
- 6 opened with "verify is registered on monolith only", which stopped being true when wave 3 landed the split arm, and cited CMakeLists.txt:1684-1760 for a block that starts around :1715; it now reads as the gate's STARTING POINT, cites the guard by name rather than a stale range, and names the block that registers the split arm
- 8's (7) marker said "done" flat, a sentence away from the midpoint tally that lists it as not done; it is now dated to wave 3 V1 and says the two sentences are about different moments
- no SHA citation touched
2026-09-22 21:36:07 -04:00
swung0x48 9e5effa9d9 [MG_IntegrationTest, docs] (P7): the spawn server does have MG_Impl - what it has not is a live context
- the verify-split block and notes/p7/verify-split.md 3.3 both said the spawn server runs no compare "because MG_Impl is not in it", which is false and sent a reader looking for a link-time gap: MobileGLServer is an executable over ServerEntry.cpp linked against the WHOLE library (top-level CMakeLists.txt), PipeFill.cpp included, so MGPipeVerifyReadHook is present and is called on every backend read
- the true mechanism: the server image's main reaches MG_Backend::InitServerRoleForSpawn() and never MG_State::Init(), so MG_State::pGLContext is null and the hook returns at its ctx == nullptr line before comparing anything
- the inproc-only conclusion is unchanged; what changes is the shape of the fix, which is a cross-process compare rather than a link edit
- CONTRACT-P7 12 records that debt by name, and the note's 6 cites it
2026-09-22 21:35:39 -04:00
swung0x48 30d3595213 [ci] (P7): the verify-split artifact keeps the green lane run, not the controls that overwrote it
- both negative-control steps re-run 12 ambient VerifySplit. entries with a knob exported, and an ambient entry writes to its own per-entry path under verify-split-logs/ which the library opens "w" - so the upload at the end of the job collected twelve CONTROL logs sitting under the LANE's names
- if: always() made it worse rather than better: a failure investigation is exactly when the overwritten half is the half somebody needs
- the arm proof is now followed by a cp -a to verify-split-logs-lane/ (plus the lane's JUnit), and the upload takes both trees - the green run the arm proof read, and whatever the controls left behind
2026-09-22 21:35:06 -04:00
swung0x48 b4f830942b [MG_IntegrationTest, ci] (P7): the verify-split arm proof is a per-entry census, not a floor
- the landing shape was a count - at least 850 armed of at least 1000 client logs - which leaves the remaining 155 unnamed: the lane could lose fifty arms, gain fifty skips, and the floor would not move
- the rule is now per entry and two-sided, the one dualblock-expected-fatals.txt already rides: each client log must carry the arming line, or belong to an entry ctest reported SKIPPED, or be named in the new Harness/verify-split-unarmed-expected.txt
- a gtest skip is the recognised no-verb marker and is NOT named, because the skip set moves with the driver - naming it would ratchet on lavapipe rather than on the comparator
- both other directions fail: a listed entry that armed, and a listed name with no client log in this run
- split_log_paths.py gains the verify-split-arming mode; the lane step gains --output-junit so a skip can be told from a pass; test_split_log_paths.py gains six controls covering the four ways the census can be wrong
- measured on the host over 1064 private client logs: 909 armed, 99 unarmed and skipped, 56 unarmed and named in five classes - caps/limits/extension advertisement 22, shader and program object surfaces 16, object state written and read back or a call rejected before it becomes a verb 14, framebuffer completeness queries 2 (DirectGLES only; DirectVulkan skips both), the poison parent that only forks 2
2026-09-22 21:35:06 -04:00
swung0x48 2424c8209d [MG_IntegrationTest] (P7): the split poison control says which role fired, not merely that someone did
- ReadChildLog concatenated the child's two role halves and the assertion searched the union, so VerifySplitPoisonOmitted. would have been just as green if the stamp the knob withheld had been the CLIENT's - which is the case it exists to rule out, the mechanism under test being the server-side verb stamp honouring MOBILEGL_PIPE_POISON_OMIT
- kOmittedPairs carries FiresOnTheServer; ReadChildLogHalves keeps Client, Server and the union apart; the split pair now asserts the Fatal in the child's SERVER half under MOBILEGL_BUILD_DISAGGREGATED, the monolith pair keeps the union because it has one role
- measured on both backends: client half 0 Fatal{UnmigratedPipeInput lines, server half 1 each - the union search could not tell them apart
- monolith PoisonOmitted. entries stay green (2/2), split VerifySplitPoisonOmitted. stay green (2/2)
2026-09-22 21:35:06 -04:00
swung0x48 e9b3f5fe04 [MG_Impl, MG_IntegrationTest, ci] (P7): the compare-at-read hook gets a negative control of its own
- MGPipeVerifyReadHook honours MOBILEGL_PIPE_VERIFY_CORRUPT: it perturbs the ORACLE after CopyField, and AGAIN after the ID-49 neutral-pack overwrite, because that overwrite replaces the pack half wholesale and CorruptStorage perturbs an array's element 0 - which is the pack half
- until now the knob only ever reached EntryCompare, so every red it could produce was the client thread's where=entry and the hook that runs on the server's apply thread at every backend read had no falsifier at all
- new entry per backend, DirectGLES/DirectVulkan.VerifySplitReadCorrupted., with CORRUPT=GetPixelStoreParameters and VERIFY_FATAL=0 - the field is in the kReadback fill mask, so the client entry compare would otherwise abort the process before the server reached the read
- the case reads the SERVER role's half of its private log alone: with the knob armed a union search over both halves would be satisfied by the arm that is not under test
- PipeStatsWindow.h gains ReadServerLogSince for that
- test.yml grows the matching control step - run the two entries, then re-read both server halves from outside for the where=read report
- measured: 2/2 green, server halves 3 and 2 lines, client halves 0 lines; revert-and-run with the two corruption blocks removed gives 2/2 red and both server halves 0 lines
- the lane goes 1070 to 1072
2026-09-22 21:35:06 -04:00
swung0x48 ca8cbe5c34 [MG_Pipe, MG_Remote, MG_Impl] (P7): the ID-49 neutral read_pixels pack has one spelling
- MG_Pipe/MGPipeTypes.h gains inline MGPipeNeutralReadPixelsPack() next to MGPPixelPackState, whose owner it is; the applier that installs the neutral state and the comparator whose oracle it is now call the same function
- MG_Remote/Server/PipeApplier.cpp read_pixels takes the constant from there instead of building it field by field
- MG_Impl/Pipe/PipeFill.cpp drops ServerReadPixelsNeutralPack, the re-typed second copy: a drift between two hand-typed spellings costs a FALSE DIVERGENCE on the server read, not a build break, which is the direction that fails quietly
- notes/p7/verify-split.md 2.2 says one spelling and why the old reasoning (do not reach into MG_Remote/Server from MG_Impl) did not survive the failure-direction argument
2026-09-22 21:35:06 -04:00
swung0x48 f4b0662d7e [docs] (P7): magma-b3 - fix round 2: the audit's anchored log, char prefix and walk stops, the server-log fatal after the ARMED check on every transport, the pull-library control, and the round's gate
- §0: the previous round's note commit gets its own row (20), and rows 21-27 for this round
- §1.5: the three audit holes the second review found and the fixtures that pin them, with
  each new fixture's reading on the previous audit and on this one (three false greens, two
  false reds, one that stays red on purpose); 14/14, tree 53 / 53 / 0
- §4.1: the server-log fatal now sits after the marker and ARMED checks and covers inproc;
  the four fail-closed demonstrations with the fatal at its new position, and the case where
  the old order reported the server-log message for a launch-failure shape
- §4.5 (new): the pull-library control had refused its own red since P6's per-role log rename
  - read from Log.h and the runner, then measured with the real ctest - and the one-place fix
- §7.2: the landed-tree line of the stale WaitForFrameSerial comment corrected; two debts
  recorded: the census sits after the replay's exit-code fatal, and B4's restart must not
  read `record` after OnSubmitsCompletedUpTo erases it
- §5: the round's gate readings on the code head
2026-09-22 21:28:02 -04:00
swung0x48 d80dc21299 [MG_Backend] (P7): comment-only - the tally header names the macro that exists, and the def says the wire arm declines, not the draw
- WireDeclineTally.h cited `MGL_WIRE_DECLINE_FALSE`, which does not exist; the statement pair
  a site actually writes is `MGL_WIRE_DECLINE_AT(Site, "why", ...); return false;`, and the
  bare-Count sites stand under a W/E of their own, which the audit holds them to
- WireDeclines.def's banner said "EVERY WAY A MAGMA WIRE DRAW CAN DECLINE", contradicted by
  the header's "a decline is not always a skipped draw" paragraph: the Tex*/Shape*/Upload*
  rows count per texture sync, reached from the descriptor resolver, attachments,
  CopyTexSubImage, mipmap generation, CopyImageSubData and readback as well as from a draw.
  The banner now says the wire arm, and carries the per-sync sentence with a pointer to the
  header
- both files are disaggregated-only (the def is included only from the guarded header), and
  the edits are comments: the pull build's .text and symbol set are unchanged (G1)
2026-09-22 21:28:02 -04:00
swung0x48 46bda06226 [scripts] (P7): the pull-library control accepts the no-client-log sentence a pull build has produced since the per-role log rename
- retrace_pull_library_control.sh required run_trace_case.cmake's "never reported resolving
  it", the marker-search sentence. Since P6's one-log-per-role rename a pull library never
  reaches that search: it has one role and writes MOBILEGL_LOG_FILE_PATH unchanged
  (output/mobilegl.log, Log.h's pull arm of RoleLogPath), the runner reads
  output/mobilegl.client.log and stops one check earlier with "the run wrote no <client log>,
  so there is no evidence the transport ever resolved" - so the control's hole-2 check
  refused its own red and the step exited 1 on every pull-library run
- measured on this tree with the real ctest, build-linux's pull library over build-split's
  frozen split library, SPLIT_TRANSPORT=inproc: control rc 1, "::error::the split retrace
  went red (ctest exit 8) ... never says the library did not resolve the transport", the
  runner's fatal at the no-client-log check. With this change the same run passes for its
  own reason on both inproc and spawn
- EVIDENCE carries both sentences as grep -F's newline-separated patterns, so the evidence
  line the red-check perturbs is unchanged; both are the transport-identity assertion
  refusing the same library for the same reason, and neither is a loader failure, a missing
  fixture, a timeout or an SSIM drop
- stub_ctest.sh gains retrace-evidence-nolog, the no-client-log fatal CMake-wrapped across
  lines the way the real one arrives; control_smoke_test.sh expects it to PASS (23/23); the
  red-check still reds "red without the transport-resolution message"
2026-09-22 21:28:02 -04:00
swung0x48 4d54027f82 [tools] (P7): a split retrace with no server log is a fatal on every transport, checked after the marker and the ARMED sentence
- the spawn server-log fatal from the previous round sat BEFORE the marker check and the
  `spawn ARMED ... pid` check, so a launch or handshake failure - no server process, no ARMED
  line - reported "the run wrote no server log" instead of the more specific "resolved, but
  carries no spawn ARMED", and sent the reader to the wrong process. It now sits just before
  the census; the tcp arm's "no forwarded server log" check, which had the same placement,
  moves with it
- the inproc row demands the server log too. Under inproc every server-role line is routed
  by thread role to the `.server.log` sink, opened lazily on the first such line, and at the
  INFO level this lane demands ServerLoop writes one when mgl-srv-apply starts - so a clean
  inproc run always has the file, and "no file" read as "0 lines" was a green Fatal{ and
  MGWIRE-FLOOR census of the role that emits both. The two census reads are unconditional
  now: the fatal above them is what makes the file exist
- fail-closed demonstrations, OpenRA DirectVulkan, the replay wrapped to damage one log after
  a real run: inproc with the server log deleted went rc 0 on the previous runner (both
  censuses printed 0) and is now a fatal at the census; spawn with the server log deleted is
  the same fatal; spawn with the server log deleted AND the ARMED line removed reports the
  ARMED message, where the previous runner reported the server-log one; spawn with only the
  ARMED line removed reports the ARMED message. Undamaged, OpenRA DirectVulkan and DirectGLES
  SPLIT and SPAWN stay green, 0 unsound lines
2026-09-22 21:28:02 -04:00
swung0x48 b0d4d730ba [tools] (P7): the audit's block walk stops at a same-indent exit and steps over a column-0 #if
- a same-indent `case`/`default:`/`break;`/`return`/`continue;`/`goto`/`throw` line between a
  log and a bare Count ends the walk: control cannot flow from that log to the Count, so it
  is not the Count's log. A `case 1: MGLOG_W; break; case 2: Count` switch with its labels at
  statement indentation, and a log above a `return` with the Count under a same-indent label,
  both read as logged before
- `#if`/`#ifdef`/`#ifndef`/`#endif` lines are passed over: a log and its Count that straddle a
  `#if MOBILEGL_BUILD_DISAGGREGATED` are one block on both sides of the preprocessor, and the
  column-0 line ended the walk for no defect. `#else`/`#elif` still end it - a log on the other
  side of one is in a different build, not this one
- four fixtures: same-indent return (red), same-indent switch (red), log across a #else (red,
  and red on the previous audit too - it pins that #else still stops), Count under its log
  across #if/#endif (green; a false red before). 14/14 fixtures; the tree stays 53 / 53 / 0
2026-09-22 21:28:02 -04:00
swung0x48 576fc09a31 [tools] (P7): the audit's strip reads a prefixed character literal as a character, not as a digit separator
- L'"', u'"', U'"' and u8'"': the `'` after an identifier character was taken for a C++14
  digit separator, the `"` inside it opened a phantom string, and the rest of the line was
  blanked - a site on that line vanished and its row read as NO SITE (fail-closed: a false red)
- CHAR_PREFIX accepts a u8/u/U/L prefix before `'` the way RAW_PREFIX does before `"`, with
  a non-identifier character (or the start of the text) in front of it, so a real separator
  (0xAB'CD, 1'000) is untouched
- self-test fixture with L'"' and u8'"' ahead of two AT sites on their own lines: rc 0; it
  was rc 1 on the previous audit. The tree has no prefixed character literal today (53 / 53 / 0)
2026-09-22 21:28:01 -04:00
swung0x48 9d5f5012c7 [tools] (P7): a site's log is a statement that begins with MGLOG_W/E, not a line that mentions one
- the wire declines audit's LOGGED pattern was unanchored, so a guarded one-liner at the
  Count's indentation (`if (g) MGLOG_W("x");`, `g ? MGLOG_W("x") : (void)0;`) counted as
  the site's own log and the site read as logged - a decline counted every time but logged
  only sometimes, which is the unlogged site the check exists to refuse
- anchored to the statement start; the block walk matches the line, not searches it
- self-test fixture "guarded one-liner log at the Count's indentation" (both shapes) reds
  with the NO_LOG sentence; it was green on the previous audit. The tree stays 53 / 53 / 0
2026-09-22 21:28:01 -04:00
swung0x48 bc3ebe7442 [docs] (P7): ID-P7-39 - origin advances to 87584d0b, F lands with fixes, and the status page says what each package's fix round is closing 2026-09-22 21:27:16 -04:00
swung0x48 87584d0b01 [docs] (P7): ID-P7-38 - B2 round 2 and the B3 fix round land with fixes, F's gate is green, E1's census is on the tree
- the two ROADMAP row segments E1 changed that did not merge onto the integrator's rows
  are re-applied by hand (host census results, P9 exception list empty, wave 3 items done)
- the running table names the fable rounds B2 r3 and B3 fix r2 and the F review
2026-09-22 21:13:16 -04:00
swung0x48 99882304f6 [docs] (P7): E1's census pages cite the rewritten commits
- the census and §7.2 pages were written against 1135c664 before the trailer rewrite; the
  citations now name its rewrite 3c80cd62 and the other remapped commits
2026-09-22 21:11:19 -04:00
swung0x48 b05406b5d9 [docs, tools] (P7): CURRENT_STAGE_STATUS §5 and the ROADMAP P3b/P4b and P7 rows mark the §7.2 re-run and the Iris census done at 1135c664, with the P3b/P4b gate "every Iris trace" now pointing at host results and an empty P9 exception list 2026-09-22 21:11:19 -04:00
swung0x48 03145fa464 [docs, tools] (P7): MEASUREMENTS §7.2 same-name re-run at 1135c664 - the 27 P5 inproc wrong answers read 24 passed / 3 named stops / 0 failed on inproc and spawn, the 1267-name lane 1051/172/37/7, and the ROADMAP debt row and CONTRACT-P7 §11/§12 lines are closed on those numbers 2026-09-22 21:09:20 -04:00
swung0x48 70bbb1591c [docs, tools] (P7): iris-census-1135c664 - all 231 Iris-corpus replays (39 cases x 2 backends x monolith/inproc/spawn) live at head, split byte-identical to monolith on 70/77 rows, texture-remint-pull never reached, so the P9 exception list is empty 2026-09-22 21:09:20 -04:00
swung0x48 bd6cd40bf8 [docs] (P7): ID-P7-37 - B2 round 2, the B3 fix round, V1 and F's first three slices are on the integration tree with a green gate, and §12 carries their recorded debts
- the trailer rewrite the user asked for is recorded (origin 1135c664 -> 3c80cd62, same trees)
- CONTRACT-P7 §12 transcribes the debt lines from magma-b2.md §6, magma-b3.md §7.2,
  verify-split.md §6 and ph-f.md §4
- the status page's gate table, landed table, running table and next steps follow
2026-09-22 21:03:51 -04:00
swung0x48 5505620e8c [docs] (P7): ph-f.md records wave 2-F - PH-7 (1)-(4) and PH-8 landed with their red-once lines and the fingerprint change, D11 / PH-2 / PH-6 / PH-1 (3)(4) not started - and the PLAN rows and CONTRACT-P7 §10 item 3 say so 2026-09-22 21:00:57 -04:00
swung0x48 ce83651349 [MG_Remote, MG_Test, ci] (P7): a Hello asking for 64 GiB windows is welcomed with the server's own terms and its session child never reaches 64 GiB, pinned in process and on the TCP supervisor 2026-09-22 21:00:57 -04:00
swung0x48 20ab14daf8 [MG_Remote, MG_Test, ci] (P7): a TCP session's data connection is bound to its authenticated control connection by a 128-bit Welcome.dataNonce instead of by arrival order, so a reordered or stale data connection is refused by name and the wire fingerprint moves with the control schema 2026-09-22 21:00:57 -04:00
swung0x48 8fa1821ab2 [MG_Remote, MG_Test, ci] (P7): one constant-time token comparison serves both server policy sites, a configured token shorter than 16 bytes refuses the listen by name, and CI builds flatc so the TCP supervisor smoke runs and must pass 2026-09-22 21:00:57 -04:00
swung0x48 8a445a3098 [docs] (P7): every cited feat/disaggregated commit is the rewritten one without the attribution trailer
- 144 commits from 785fed0b0 onward were rewritten to drop the Co-Authored-By trailer the
  project forbids; origin/feat/disaggregated moved from 1135c664 to 3c80cd62 (same trees)
- the old->new map is ~/w7/logs/sha-map-20260922.txt; every 7..40-hex citation in the
  Disaggregated docs and the MG_Remote contracts that matched an old SHA is replaced by the
  same-length prefix of its rewrite; no other content changes
2026-09-22 20:51:15 -04:00
swung0x48 a7a25a75b5 [docs] (P7): magma-b3 - the note describes the fix round: the audit's strip, block walk and self-test, the spawn server-log fatal, and the round's gate
The fix round's commits changed what §1.5 says the audit checks and what
§4.1 says the runner reads, so the note is brought up to the tree:

- §0: rows 11-19 for the fix round.
- §1.5: the three holes (a site in a comment or string, the branch-blind
  8-line window, a .def row with a trailing comment), what the audit does
  now, and the eight --self-test fixtures run first in the CI step.
- §4.1: the spawn server-log fail-closed and its demonstration (runner at
  aab01275 rc 0 on a run whose server log was removed; this runner rc 1,
  the CMake Error quoted). inproc also writes mobilegl.server.log and is
  still read if(EXISTS); only spawn was tightened, as the review asked.
- §5: the fix round's gate on package tree 88d1a913, pre-merge: split
  build rc 0; G1 rc 0 .text 0xa52203 nm 0/0; census 79; audit self-test
  8/8, tree 53/53/0; ratchet unchanged at 186; parity rc 0; unit
  2420/2420, integration-magma-split 90/90, integration-magma-spawn 69/69;
  OpenRA DirectVulkan SPLIT / SPAWN ssim 1.000000, 0 px, 0 MGWIRE-FLOOR.
2026-09-22 20:49:46 -04:00
swung0x48 aa8ef34641 [docs] (P7): magma-b3 - §7.2 records that WaitForFrameSerial can answer true for a serial the clamped floor has not reached, for the integrator to copy into §12
Fable review of the B3 rework, note for CONTRACT-P7 §12. On the
disaggregated build VulkanRenderer::WaitForFrameSerial waits the FIRST
in-flight record with frameSerial >= serial, calls OnSubmitsCompletedUpTo
and returns true. With the §3.1 clamp, a later record carrying the same
serial (S2, the Present submission - the two-submissions-per-serial shape
of §2.2) keeps the floor below `serial`, so the function returns true
while IsFrameSerialComplete(serial) is false. The only caller,
WaitForTimerQueryResult, re-checks IsTimerQueryResultReady, so today it
degrades to a spurious timeout, not a wrong read.

The fix belongs to B4 and is not made here: after OnSubmitsCompletedUpTo,
return only if IsFrameSerialComplete(serial), else restart the scan (the
call erases from the vector being iterated, so restart, never continue).
2026-09-22 20:49:45 -04:00
swung0x48 22c157cd8b [docs] (P7): magma-b3 - the stale WaitForFrameSerial comment is cited by function, with its line on both trees
Fable review of the B3 rework, nit: §7.2 cited the comment as
VulkanRenderer.cpp:13400-13403, which is the package tree's line; on the
landed tree pipe@6207c9c9 it is 13530-13533. The bullet now names the
comment by where it sits (above the fence-wait loop in
VulkanRenderer::WaitForFrameSerial) and gives both lines, the function name
being the reference.
2026-09-22 20:49:45 -04:00
swung0x48 aa6fe6d9da [docs] (P7): magma-b3 - the gate numbers say which tree they were read on, and the landed tree has its own row
Fable review of the B3 rework, should-fix: §5's table and prose (unit 2420,
integration-split 226, magma 90/69/70, full 527, ratchet 186) were read on
the package tree before the merge, but read as if they described what
landed; and the quoted parity output ("1 entry not registered on the tcp
arm by name") is the package script's, which did not land.

- The initial gate is labelled package tree 0d425499, pre-merge; the
  review-round gate package tree 774bc17f, pre-merge.
- A landed-tree column/paragraph from the integrator's gate on
  pipe@6207c9c9: unit 2429 / integration-split 304 /
  integration-magma-{split,spawn,tcp} 96/75/71 / full 535; G1 rc 0 .text
  0xa52203 nm 0/0; census 79; ratchet 173; parity rc 0; audit 53 rows 53
  sited 0 unlogged; OpenRA DirectVulkan inproc + spawn retrace ssim
  1.000000, 0 px, 0 unsound-serial-complete lines. The landed lane counts
  include other packages' entries.
- The parity quote is replaced by the landed wording ("N server-env-knob
  entrie(s) excluded from the TCP comparison only"), and §4.2 / §5 name
  the landed list MAGMA_SERVER_ENV_KNOB_NO_TCP (compare_arms(no_tcp=),
  CMake mglItestServerEnvKnobArm), into which the integrator folded
  MAGMA_NOT_OVER_TCP.
2026-09-22 20:49:45 -04:00
swung0x48 7e51860435 [MG_Backend, docs] (P7): a wire decline's line says decline, not draw, because the texture rows count per sync
Fable review of the B3 rework, nit: MGL_WIRE_DECLINE_AT logged "Magma wire
draw declined [Site]" from every site, but the Tex*, Shape* and Upload*
rows are reached through SyncTextureResourceByHandle, whose callers are not
only the draw/dispatch texture preparation: the descriptor resolver,
framebuffer attachments, CopyTexSubImage, mipmap generation and
CopyImageSubData each Fatal right after a failed sync, and texture
readback falls back to the unbacked shadow. Those counters are per sync,
not per draw - a draw sampling three unsyncable textures counts three, a
failed attachment sync counts once and is then a Fatal.

- WireDeclineTally.h: the line is "Magma wire decline [Site]: ...", and
  the macro's comment says the texture rows are per sync and names the
  callers. Disaggregated-only header; no test or script matches the old
  text (git grep).
- magma-b3.md §1.5: says the counters are per sync and that the log lines
  quoted in §1.3 / §1.5 carry the old wording.
2026-09-22 20:49:45 -04:00
swung0x48 4196f0e902 [MG_Backend] (P7): the image-flags decline's comment says it was one of five Shape* exits with no shipped line, not the one
Fable review of the B3 rework, nit: the comment over the ShapeImageFlags-
Unsupported site called it "the one Shape* exit that left no line on the
builds that ship". At 0d425499 SyncWireTextureShape had five such exits:
ViewOf (bare return false), NoLevels (bare return false), preserve-flush
(bare return false), image flags (MGLOG_D, compiled away in Release) and
preserve-copy (bare return false). The comment now names all five.

Comment-only, inside the disaggregated-only SyncWireTextureShape; pull
build unaffected.
2026-09-22 20:49:45 -04:00
swung0x48 7ae1c11de0 [tools] (P7): a spawn retrace with no server log is a fatal, not a clean census
Fable review of the B3 rework, should-fix: run_trace_case.cmake read the
server log only `if(EXISTS ...)`. Under spawn the server process writes its
own file, and the lines the runner counts out of it - the applier Fatal{
lines and the MGWIRE-FLOOR unsound-serial-complete line, which only the
server's VulkanRenderer::OnSubmitsCompletedUpTo emits - are nowhere else,
so a server that died before its first line read as "0 lines" and went
green. The client log was already fail-closed, and so was the tcp arm's
forwarded server log; plain spawn was not.

The plain-spawn branch of the transport block now FATALs when
mobilegl.server.log does not exist, beside the tcp arm's identical check.

Demonstration: the SPAWN ctest command line with TRACE_REPLAY_EXE replaced
by a wrapper that runs the real replay and then deletes
output/mobilegl.server.log (541 lines):
  runner at aab01275 -> rc 0, "Fatal{ lines across ... : 0",
    "MGWIRE-FLOOR unsound-serial-complete lines: 0" (green, half unseen)
  this runner        -> rc 1:
    CMake Error at run_trace_case.cmake:321 (message):
      OpenRA DirectVulkan: MOBILEGL_TRANSPORT=spawn but the run wrote no
      .../output/mobilegl.server.log, so the server role's Fatal{ and
      MGWIRE-FLOOR lines cannot be counted.  A spawn retrace with no
      server log cannot be counted as a clean one.
MobileGLTraceReplay.OpenRA.DirectVulkan.SPLIT / .SPAWN unmodified: both
pass, ssim 1.000000, 0 px, 0 MGWIRE-FLOOR lines.
2026-09-22 20:49:45 -04:00
swung0x48 266d6a32f8 [tools] (P7): the wire declines audit carries a self-test whose fixtures must red, and the CI step runs it before the tree audit
Fable review of the B3 rework, should-fix: the audit had no negative
control, so a green run could not say it would have been red. It now takes
--self-test, which writes each inline fixture as a one-file renderer tree in
a temp dir, runs the audit on it and checks both the rc and the line the
audit must print (so a red is red for the fixture's reason):

  comment-only site (//)                -> rc 1, "has NO SITE"
  comment-only site (/* */)             -> rc 1, "has NO SITE"
  string-literal site                   -> rc 1, "has NO SITE"
  other-branch log                      -> rc 1, "with no MGLOG_W/E in its own block"
  nested-branch log                     -> rc 1, same
  log only in a /* */ comment above     -> rc 1, same
  nearest log is MGLOG_D                -> rc 1, same
  good sites (AT form, bare Count under
    its own MGLOG_E_ONCE, def row with a
    trailing comment)                   -> rc 0, "0 unlogged, 0 unknown"

8/8, rc 0. Mutants: with the strip disabled 5 of 8 fail (the four
comment/string reds go green and the good fixture's commented row is
lost); with the old 8-line window in place of the block walk 2 of 8 fail
(other-branch, nested-branch). Against the audit as it stood at aab01275,
six of the seven red fixtures passed and the good one failed.

.github/workflows/test.yml: the "Wire declines audit (B3)" step runs
--self-test and then the tree audit, the way the next step runs
trace_cases.py --self-test-transport-matrices before it computes matrices.
2026-09-22 20:49:45 -04:00
swung0x48 78269b2677 [tools] (P7): a bare decline count is logged only by a warning-or-error line in its own block, and a def row with a trailing comment is still a row
Fable review of the B3 rework, should-fix: the 8-lines-above MGLOG_W/E window
was branch-blind - a Count() in `if (x == 2) { ... }` whose only log sat in a
preceding `if (x == 1) { ... }` passed, as did a log in a nested branch just
above. All 12 bare sites in the tree are under their own line today, so this
was a gate hole, not a present defect.

has_own_log() walks upwards from the Count line (still capped at 8 lines):
blank and deeper-indented lines (a log's continuation lines, a nested block)
are passed over, only a line at the Count's own indentation can be its log,
and the first line indented less than the Count - the block's opening, a
sibling's `} else {` - ends the walk.

WireDeclines.def is now read through the same comment/string strip, and ROW
tolerates spaces inside the parentheses: `MGL_WIRE_DECLINE(Name) // note`
used to fail `\)\s*$`, dropping the row, and its real site was then
reported as naming no row.

Tree: 53 rows, 53 with sites, 53 sites, 0 unlogged, rc 0 (unchanged).
Fixtures the previous commit passed and this one reds: other-branch log,
nested-branch log (rc 0 -> 1); a def row with a trailing comment plus its
site (rc 1 -> 0).
2026-09-22 20:49:45 -04:00
swung0x48 88599fc6f6 [tools] (P7): the wire declines audit matches sites and logs on code alone, so a commented-out or quoted site is no site
Fable review of the B3 rework, must-fix: a row was "sited" by any line the
AT_SITE / COUNT_SITE regex matched, including a // comment, a /* */ block or
a string literal - `// MGL_WIRE_DECLINE_AT(GhostRow, "...");` satisfied the
audit for a row nothing counts, and an MGLOG_W inside a comment above a bare
Count() satisfied the log rule.

wire_declines_audit.py now blanks //...$ (with backslash-newline
continuation), /* ... */, "..." and '...' (raw strings R"d(...)d" and
C++14 digit separators handled) out of each source before any match. Every
character but the newline becomes a space, so the text keeps its length and
line count and each report still cites the real line.

Tree: 53 rows, 53 with sites, 53 sites, 0 unlogged, rc 0 (unchanged).
Fixtures that the previous audit passed and this one reds (rc 1): a
comment-only site (//), a comment-only site (/* */), a string-literal site,
a log only in a /* */ comment above a bare Count. The branch-blind window
and the .def trailing-comment row are the next commit.
2026-09-22 20:49:45 -04:00
swung0x48 c997ba9425 [docs] (P7): magma-b2.md carries review round 2 - where the flipped case asserts, the padded stride, one decision per blit, MsFlip1 - two new debt lines, and gates for this tree and the landed one
- §2.9 (new): the lane-marker/backend skips and their verified Skipped lines, the 4-byte row stride and its validation-layer red-once, the whole-call shape decision (the round-1 partial effect: a scaled or X-mirrored COLOR|DEPTH blit wrote its colour and then errored on its depth) with its red-once, the split once-sites, the empty-side decline, MsFlip1 and the exact-one parity tail check, each red-once as run.
- §2.6: MsFlip goes red on a device whose renderbuffer and texture D24S8 map to different VkFormats (cross-format Y decline); §5 item 2 says to check it on the Redmi.
- §2.8: the round-1 bufferRowLength=0 note and the per-aspect decline are marked superseded by §2.9.
- §6: item 1 cites VulkanRenderer.cpp:10066 (the vkCmdBlitImage; :10060 is the vkCmdResolveImage) and states the scale partial effect as closed; item 5 states the X/cross-format-Y partial effect as closed and that its INVALID_OPERATION is a rule I (a) refusal of a legal blit; item 4 counts six knob entries; new item 6 (monolith DirectVulkan flipped/scaled multisample depth blit, VulkanRenderer.cpp:9594/:9608, a wire-vs-monolith divergence where the wire arm is the correct one, P13/G1-bound) and item 7 (Espryt flipped/scaled multisample depth resolve, P3b/P4b); stale WireFramebuffer.inc line numbers refreshed.
- §7 is two labelled rows: this tree (p7/magma-b2@4442ea6d, measured) and the landed tree (pipe@3b66d1b1, integrator-measured), instead of numbers pinned to 5d896170, which is not an ancestor of the landed tree.
2026-09-22 20:49:45 -04:00
swung0x48 4db17d58c6 [MG_Backend, MG_IntegrationTest, scripts] (P7): a declined multisample depth/stencil shape is decided once before any aspect runs, so an erroring blit writes nothing, and the shader resolve arm gets a flipped knob twin
- BlitWireFramebuffers runs colour first. A scaled COLOR|DEPTH blit out of a multisample framebuffer scaled and wrote its colour and then recorded INVALID_OPERATION on its depth, and an X-mirrored one mirrored its colour and left its depth alone - partial effects GL 18.3.1 does not allow. The shape decline (scale, X mirror, cross-format Y mirror) moves into DeclineWireDepthStencilResolveShape, which BlitWireFramebuffers asks for every depth/stencil aspect that would take the multisample resolve arm before any aspect runs; ResolveWireDepthStencil asks it too for any other caller. Default read/draw framebuffers are left to the lambda (a default read is never multisampled; a default draw behind one is the region Fatal, and resolving it early would move its write index).
- The X decline and the cross-format Y decline get their own MGLOG_E_ONCE sites; one site used to cover both, so after the first X decline a cross-format Y decline logged nothing.
- An empty rectangle on one side only (64x48 onto 0x48) is a size mismatch and now takes the scale decline (INVALID_OPERATION) instead of returning silently; empty on both sides stays a silent no-op.
- DirectVulkan.{Split,Spawn}.MsFlip1. forces MGITEST_MAGMA_FORCE_SHADER_DEPTH_RESOLVE=1, so the shader arm min-corner rectangle under a flip runs on this host; split+spawn only and named in MAGMA_SERVER_ENV_KNOB_NO_TCP.
- spawn_lane_parity.py: every server-env-knob tail must match exactly one split key, so a dead or over-broad tail fails the gate instead of riding silently.
- The flipped-resolve case gains a COLOR|DEPTH scale leg (error, colour untouched, depth untouched) and an empty-destination leg (error, depth untouched).
- Red-once, each on split + spawn (+ tcp MsFlip.) and restored: pre-pass disabled -> all five MsFlip/MsFlip1 entries fail "the declined COLOR|DEPTH scale must not have written its colour either" (read {255,0,0,255}, want {0,0,255,255}); round-1 silent empty-side return -> all five fail the empty-destination leg (error 0, want 1282); shader arm handed sx0/sy0 instead of the min corner -> MsFlip1 fails on both arms ("768 of 768 depth values differ from 0.75", "928 of 3072 stencil values are wrong; first at (20, 4): got 255, want 99") while MsFlip. stays green; a dead ".MsFlip2." tail -> parity rc 1 "matches 0 split entrie(s)".
- The split once-sites have no lane red-once: the cross-format Y decline needs a device whose renderbuffer and texture D24S8 map to different VkFormats, which lavapipe is not.
2026-09-22 20:49:45 -04:00
swung0x48 1e46a67091 [MG_Backend, MG_IntegrationTest] (P7): the multisample depth/stencil copy-out pads its row stride to 4 bytes, so a narrow flipped stencil resolve puts every row region on a legal offset
- ResolveWireDepthStencil moved the resolved aspect through a tightly packed buffer and, for a vertical mirror, started one region per row at row * width * texelSize. A depth/stencil buffer->image copy needs every bufferOffset on a multiple of 4 (VUID-vkCmdCopyBufferToImage-dstImage-07978), so a stencil aspect (1-byte texel) whose width is not a multiple of 4, or a D16 aspect with an odd width, was illegal on every row after the first; scissor clipping produces such widths.
- paddedWidth = width rounded up to 4 / texelSize texels; the buffer is paddedWidth * height * texelSize, both copies name it as bufferRowLength, and each row region starts at row * paddedWidth * texelSize.
- The flipped-resolve case gains a stencil leg: stencil bands 11/99 in the multisample source, a 29x32 flipped STENCIL_BUFFER_BIT blit from (5,8) to (20,4), and every stencil value of the destination checked against the mirrored band or the untouched prime.
- Red-once (split + spawn, under VK_LOADER_LAYERS_ENABLE=*khronos_validation): lavapipe does not enforce the VU and the stencil bytes came back right on the unpadded code, so the red is the validation layer: 10 x "pRegions[1].bufferOffset (29) must be a multiple 4 if using a depth/stencil format (VK_FORMAT_D24_UNORM_S8_UINT)" on each arm before, 0 validation errors of any kind after.
2026-09-22 20:48:41 -04:00
swung0x48 0019643406 [MG_IntegrationTest] (P7): the flipped multisample resolve case asserts on the magma wire arm only and skips the monolith and espryt entries by name
- The suite is discovered whole on the monolith DirectVulkan arm and on every DirectGLES arm, and the case was red on all of them: the monolith refuses a flipped depth blit and scales a multisample one GL calls INVALID_OPERATION, and Espryt writes nothing for the flip and raises nothing for the scale.
- A backend gate first (Gl().BackendName() != "DirectVulkan", Espryt reason, P3b/P4b), then the lane marker (SplitLane::IsSplitLane(), monolith reason, P13/G1-bound), so a DirectGLES entry names the backend that owes the fix whichever lane it sits in.
- DirectVulkan.{Split,Spawn,Tcp}.MsFlip. stay armed (all carry MGITEST_SPLIT_LANE=1) and pass.
- Verified: DirectVulkan.DepthStencilReadbackMatrixScenario.AFlipped* and DirectGLES.{,ForcedDepthStencilEmulation.}DepthStencilReadbackMatrixScenario.AFlipped* report Skipped with those reasons.
2026-09-22 20:48:41 -04:00
swung0x48 1b9aa65145 [docs] (P7): the pipe commits the status page and ID-P7-35/36 cite are the ones without attribution trailers
- the sixteen unpushed integration commits after 1135c664 were rewritten to drop the
  Co-Authored-By trailer the project forbids (commit-style rule, user 2026-07-15 and
  2026-09-09); their SHAs changed, so every citation of the old SHAs is remapped here
- no content change beyond the SHA text
2026-09-22 20:42:30 -04:00
swung0x48 542c5838e2 [docs] (P7): ID-P7-36 - the B3 rework lands with fixes, one knob-exception list survives the merge, and the OpenRA split retrace with its new red condition is part of every pipe gate
The status page's exit-gate table now says item (7) verify x split is reported and
on pipe pending its gate and review, and the running table names the B3 fix round
and the V1 review.
2026-09-22 20:38:59 -04:00
swung0x48 75858bfd9c [docs] (P7): notes/p7/verify-split.md records exit gate 2 met on the host - verify x {monolith, split(inproc)} green on both backends, both controls red once on the split arm, 8 verify traces x DirectVulkan x inproc at zero divergences - and CONTRACT-P7 §6 and §8 (7) say so 2026-09-22 20:35:15 -04:00
swung0x48 38bfdf6d63 [MG_IntegrationTest, MG_Pipe, ci] (P7): VERIFY_CORRUPT and POISON_OMIT each go red on the split arm, the poison through the server's own stamp with the ReadPixels:GetPixelStoreParameters pair, and verify stays inproc-only because its two arms share only that process 2026-09-22 20:34:50 -04:00
swung0x48 64a2f4a300 [MG_IntegrationTest, MG_Pipe, ci] (P7): every verify case runs on the split (inproc) arm of both backends under integration-verify-split, armed per entry, and the comparator's oracle for the pack half inside the server's read_pixels window is the ID-49 neutral pack 2026-09-22 20:34:50 -04:00
swung0x48 e8678dfad9 [MG_Test] (P7): the respecify-scope unit helpers are compiled only where the applier they read is, so the pull build builds again 2026-09-22 20:34:50 -04:00
swung0x48 a1d17566f7 [MG_Pipe, MG_Test] (P7): the respecify scope pin asks that a per-level producer cover the range it declares, so verify and split can be in the same build 2026-09-22 20:34:50 -04:00
swung0x48 75a64ce0da [tools, docs] (P7): the parity script keeps one server-side-knob list after the B3 rework merge, and magma-b3.md names it
The B3 rework's parity hunk and B2's round-1 hunk described the same asymmetry
under two names; the merge kept B2's compare_arms(no_tcp=) and added .StaleSerial.
to MAGMA_SERVER_ENV_KNOB_NO_TCP. One line of the package tree's other mechanism
(`not_on_arm = not_on_arm or {}`) survived outside the conflict hunks and would
have been a NameError on the first run; it goes, and the note's two mentions of
MAGMA_NOT_OVER_TCP now name the list that exists.
2026-09-22 20:25:48 -04:00
swung0x48 3f92d0f109 [docs] (P7): magma-b3 - the review round: every decline row is sited, the floor has a lane, the stamp is a probe, and the tcp arm carries no knob it cannot see
Records review ID-P7-34's four should-fix items as landed:
- §1.5: the .def audit (52 rows / 40 sited / 10 unlogged, rc 1 -> 53 / 53 /
  0, rc 0), the per-function distribution, and the forced ShapeNoVkFormat
  red; §1.2's "32 exits" is corrected against the tree.
- §3.2: the submission term is a probe and a defence, not the fix - the
  stamp can name S1 while the draw rides S2 through the preserve flush -
  and the :273 early return and :432 re-stamp.
- §4.1: the MGWIRE-FLOOR red condition in run_trace_case.cmake, with the
  red (clamp deleted: 26 lines, ssim 1.0, FAIL on SPLIT and SPAWN) and the
  green.
- §4.2: StaleSerial. on split and spawn only; MAGMA_NOT_OVER_TCP.
- §5: the review-round gate sweep; integration-magma-tcp 71 -> 70.
- §7.2 (for CONTRACT-P7 §12): the disaggregated monolith arm is now
  sounder than the pull build on mid-frame flushes (OnSubData reads the same
  floor via IsResourceBusy; the pull build keeps the latent bug under G1 /
  P13), and the stale comments at WaitForFrameSerial and
  CollectAllDeferredReleases; the reviewer's verdict on the clamp and on B4.
2026-09-22 20:25:11 -04:00
swung0x48 80b91346d1 [tools] (P7): a split retrace reds on any MGWIRE-FLOOR unsound-serial-complete line, so the frame-serial floor clamp has a lane
Review ID-P7-34, item 2. The primary B3 fix - the clamp in
VulkanRenderer::OnSubmitsCompletedUpTo that raises the completed-frame-
serial floor only to a serial no in-flight submission still carries - had
nothing behind it: deleting the clamp loop left every lane green and the
OpenRA picture gold on lavapipe (the host's barriers hide the tear). Its
only observable was the per-event MGLOG_W "MGWIRE-FLOOR
unsound-serial-complete".

run_trace_case.cmake now reads that line out of both roles' logs on every
split run, exactly as it already reads Fatal{: the count is printed pass
or fail and any nonzero count is a FATAL_ERROR. The R-16 counter (26 -> 0
on the OpenRA retrace) is thereby a gate.

Red/green on MobileGLTraceReplay.OpenRA.DirectVulkan.SPLIT / .SPAWN:
  clamp loop deleted -> both FAIL at ssim 1.000000 / 0 px:
    MGPipe split: OpenRA DirectVulkan transport=inproc, MGWIRE-FLOOR
    unsound-serial-complete lines: 26
    [mgl-srv-apply/WARN]: MGWIRE-FLOOR unsound-serial-complete #1: advancing
    the completed frame-serial floor to 3, which submission 3 (serial 3)
    still carries
    (spawn: identical, 26)
  clamp restored -> both PASS, 0 lines.
2026-09-22 20:25:11 -04:00
swung0x48 cba0c3a110 [MG_Backend, MG_IntegrationTest] (P7): the busy predicate's comments name the floor as the guarantee and the submission term as a probe, and the term is consulted and re-stamped wherever the serial is
Review ID-P7-34, item 3. 02bb5789 presented the submission term as the fix
and still carried the every-8th-drain mechanism that magma-b3.md §2.4
refutes (the probe saw one successful drain and zero boundary works across
the OpenRA replay). The term is not sound alone: lastUseSubmitIndex is
stamped at AcquireWireSlice, but a mid-draw flush (BindProgramUniformBuffers
-> SyncWireTextureShape's preserve path -> FlushWirePendingCommandsFor-
TextureUpdate) can submit the stamped index without the draw, so the stamp
names S1 while the draw rides S2. The guarantee is the floor in
OnSubmitsCompletedUpTo; the term is a defence and a device probe.

- WriteWireBuffer's block comment, the WireBufferResource field comment,
  the AcquireWireSlice stamp comment, the StreamedBufferSubData CMake
  comment and the F1WireScenario case comment are rewritten on that
  footing; the drain story is gone. The drain probe's comment now says it
  refuted the drain as the mechanism.
- WaitForWireBufferHostAccess's early return also requires the submission
  term complete, so a caller with the serial term idle cannot skip a wait
  the term demands.
- ReadWireBuffer re-stamps lastUseSubmitIndex beside lastUseSerial after
  its wait, preserving the pending draw's reservation in both halves.

All disaggregated-only code. OpenRA DirectVulkan SPLIT / SPAWN retraces
ssim 1.000000, 0 px, zero MGWIRE-FLOOR lines; integration-magma-split 90/90,
-spawn 69/69, -tcp 70/70.
2026-09-22 20:25:11 -04:00
swung0x48 b0c6af3168 [MG_IntegrationTest, tools] (P7): the stale-serial red-once is registered on split and spawn only, and the parity gate names it with the other server-side knobs
MGITEST_MAGMA_FORCE_STALE_BUFFER_SERIAL is read by the server. The tcp
arm's server is the lane fixture (scripts/ci/tcp_server_fixture.py),
started once from its own environment, so an entry-level ENVIRONMENT never
reaches it: DirectVulkan.Tcp.StaleSerial.* ran the unforced case under a
name that claimed the knob and could never red (review ID-P7-34, item 1).

The entry is now registered only when the arm is Split or Spawn, and
spawn_lane_parity.py grows a second named asymmetry beside
MAGMA_INPROC_ONLY: MAGMA_NOT_OVER_TCP, key prefixes the reference arm has
and the tcp arm does not by name. It comes off the reference for the tcp
arm alone, so the same entry missing from spawn is still an error. B2 adds
its own knob entries to the same tuple in its tree.

integration-magma-tcp 71 -> 70 entries (68 cases + the two fixture
entries); split 90, spawn 69 unchanged. spawn_lane_parity.py build-split
rc 0; tcp_lane_tools_test.py 6/6.

Integrator merge onto pipe: B2's round-1 mechanism (`mglItestServerEnvKnobArm` in
CMake, `compare_arms(no_tcp=)` in the parity script) already covers this shape, so
`.StaleSerial.` joins MAGMA_SERVER_ENV_KNOB_NO_TCP instead of the separate
MAGMA_NOT_OVER_TCP the package tree introduced.
2026-09-22 20:25:11 -04:00
swung0x48 b8d10d0730 [MG_Backend, tools] (P7): every WireDeclines.def row has a site and every site a warning-or-error line, and a tree-only audit fails the job when one does not
The B3 review (ID-P7-34) counted 12 of the 52 rows with no site at all -
TexViewFormat, TexNoApplierRecord, TexRecordDeadOrStale and the nine Shape*
rows were declared and never counted - and ten sites that counted with no (MGLOG_D included)
line, so "a name, a tally and a decline line" did not describe the tree.

- SyncWireTextureShape: all nine Shape* exits now count; the ViewOf exit
  and the preserve-flush exit were bare `return false` and are now
  MGL_WIRE_DECLINE_AT; the image-flags exit logged only MGLOG_D, which a
  Release build compiles away, and is promoted likewise.
- The preserve-copy exit was the one Shape exit with no row: appended as
  ShapePreserveCopyFailed (rows are append-only), 53 rows.
- SyncTextureResourceByHandle / UploadPendingWireLevels: the count-only
  sites without a log of their own are promoted to MGL_WIRE_DECLINE_AT;
  the ones already under a W/E _ONCE line keep that line and gain the count.
- scripts/ci/wire_declines_audit.py: fails on a row with no site, a site
  naming no row, or a bare Count() with no MGLOG_W/E within 8 lines above
  it. WireDeclineTally.h (the macro's definition) is not a site. Wired into
  the job that runs the fatal census.

Audit before: 52 rows, 40 with sites, 10 unlogged under the W/E rule (rc 1). After: 53 rows,
53 with sites, 0 unlogged (rc 0). Forced red (ShapeNoVkFormat made true for
1x1 textures, reverted): Split.Fm.F1WireScenario.VertexIdSampler... fails on
pixels with ShapeNoVkFormat=2 / TexShapeSync=2 in the shutdown tally.
2026-09-22 20:24:00 -04:00
swung0x48 613ed05b98 [docs] (P7): ID-P7-35 - the B2 rework's first round is sent back, because its new case also lands on the monolith and Espryt entries and its per-row copy-out needs an aligned stride
The status page and the ROADMAP P7 row now say what is in flight after the six
packages were relaunched on claude-opus-5-5: B2 rework round 2, B3 rework, M2, V1,
F and E1. pipe stays at 1b5d97bd and is not pushed until round 2 lands.
2026-09-22 20:21:23 -04:00
swung0x48 cc3c18c578 [scripts, docs] (P7): magma-b2.md counts the baseline as 73 + 11 + 2 + 2, carries the review round's §12 lines and gates, and the ratchet baseline names the tree that measured 173
ID-P7-31 review, note 3 and nits.

  * §0's preamble: the +4 over package B's 84/63/65 is not "package A landing after
    B" - A landed BEFORE B and is +2 (magma-a.md:151-153); package C is the other +2
    (magma-c.md:59-61). 73 + 11 (B) + 2 (A) + 2 (C) = 88/67/69.
  * §2.3 / §2.5: MsResolve1 is split+spawn only; the day-of +3 on tcp is marked as
    superseded.
  * §2.7 (the server-side knob has no tcp arm, measured) and §2.8 (flip mirrored,
    scale declined, X / cross-format mirror declined; three red-once lines).
  * §6: the lines for CONTRACT-P7 §12 - the scaled second leg is unclipped
    (VUID-vkCmdBlitImage-dstOffset-00248/00249, shared with the monolith);
    MagmaWireFatal("multisample-resolve-region") for a source rect outside the read
    buffer, where GL promises undefined values rather than an error; the 1->N
    identical-rect decline declines a legal shape (18.3.1 sample replication) because
    the monolith's own observable is a VU violation; the tcp lane cannot carry a
    server-side knob; X and cross-format Y mirrors of a depth/stencil resolve decline.
  * §7: the review round's gates - magma 94/73/70, full-split 525, unit 2423, parity
    RC 0 naming the five MAGMA_SERVER_ENV_KNOB_NO_TCP entries, census 79, ratchet 173,
    G1 .text 0xa52203 with 0/0 symbol drift, @P7 19 comment lines.
  * link_ratchet_baseline.txt: `base commit` points at 6af422b3, the tree that
    measured 173, not 4bee1313.

The "@P7 = 24 lines" nit needed no edit: the note already reads 19 at 6af422b3, which
is what the tree measures.
2026-09-22 20:05:28 -04:00
swung0x48 645dd96f69 [MG_Backend, MG_IntegrationTest] (P7): a flipped multisample depth or stencil resolve lands mirrored and a scaled one declines, and neither ends the session
ID-P7-31 review, should-fix 2.

ResolveWireDepthStencil required sx1 > sx0, sy1 > sy0 and identical signed spans, and
took the session with Magma:multisample-depth-resolve-region otherwise - so
glBlitFramebuffer(0, h, w, 0, 0, 0, w, h, COLOR|DEPTH) out of a multisample framebuffer
resolved its colour correctly and then died on its depth. GL 4.6 core 18.3.1 asks for
identical DIMENSIONS, not identical corners.

  * spans are signed; everything downstream (renderArea, the shader arm's rectangle, the
    buffer copy offsets, BlitDepthAcrossFormats' corners) works off the min corner;
  * a SCALE is INVALID_OPERATION in 18.3.1 and now declines by rule I (a) with the colour
    arm's three lines (MGLOG_E_ONCE, RecordTextureCopyError, return);
  * a same-format Y mirror is done in the copy-out: one VkBufferImageCopy per row,
    buffer row r written to destination row dyMin + height - 1 - r;
  * an X mirror (a region per texel is not a copy) and a Y mirror across two formats
    (BlitDepthAcrossFormats, shared with the monolith, has no flip) decline by name.

The region Fatal stays for a rectangle outside either image.

Scenario DepthStencilReadbackMatrixScenario.AFlippedMultisampleResolveMirrorsTheBandsAnd-
AScaleDeclines, registered as MsFlip. on all three magma arms (no knob): unequal depth
bands, COLOR|DEPTH in one flipped call with colour and depth asserted the same way up,
then a scaled depth blit that must be INVALID_OPERATION and leave the destination alone.

Red once (R-16, executed and reverted, split and spawn each):
  * the old guard restored -> both arms die by
    Fatal{UnmigratedVerb, "Magma:multisample-depth-resolve-region"};
  * the scale decline turned back into that Fatal -> the flip half passes, both arms die
    at the scale;
  * the per-row copy-out bypassed (if (!mirrorY) forced true) -> both arms live and fail
    "flipped resolve: the destination's bottom band: 768 of 768 depth values differ from
    0.75; first bad value 0.25" (and the top band the other way).
2026-09-22 20:05:28 -04:00
swung0x48 9fed478dfd [MG_IntegrationTest, scripts] (P7): a server-side magma test knob is registered on split and spawn only, and the parity gate names the five entries tcp cannot carry
ID-P7-31 review, should-fix 1.

Every MGITEST_MAGMA_FORCE_* knob is read BY THE SERVER (the getenv readers in
WireFramebuffer.inc run on the apply thread), and a ctest ENVIRONMENT property only
reaches the process ctest starts - the client. Inproc runs the server in that process
and ServerSpawn.cpp copies ::environ into the spawned child, so both arms see the knob;
the tcp arm's server is the lane-wide TcpServer.Start fixture, started once from
scripts/ci/tcp_server_fixture.py's own os.environ, and nothing a per-case property says
reaches it. The tcp copy of each knob=1 entry ran the DEFAULT arm under a name claiming
otherwise.

Measured, not deduced: a temporary fatal in each of the three readers killed all five
knob=1 entries on split and on spawn and left all five green on tcp.

So ShaderMip1, ShaderMip2, DepthMip, DefaultBlitShape1 and MsResolve1 are registered on
split+spawn only (the first four are package B's pre-existing vacuity of the same
shape). The knob=0 twins DefaultBlitShape0 / MsResolve0 keep their tcp entries: zero is
the default and means the same thing on every arm.

spawn_lane_parity.py names the exception, following ID-P7-14's MAGMA_INPROC_ONLY:
MAGMA_SERVER_ENV_KNOB_NO_TCP, applied through a new compare_arms(no_tcp=) that drops the
keys from the tcp comparison ONLY - folding them into inproc_only would stop requiring
them on spawn, which is where they do most of their work.
2026-09-22 20:05:28 -04:00
swung0x48 3c80cd62fe [docs] (P7): ID-P7-34 - OpenRA passes 27/27 on the phone; CURRENT_STAGE_STATUS.md is the live stage page and the ROADMAP rows say where P7, Ph and P3b/P4b stand
p7w5 (B3 on top of B2 and D3) renders OpenRA to the golden on every arm
the acceptance asked for - inproc and spawn, run-ahead and lockstep,
cold and warm, FIF=8 and the dump-armed variant that failed 0/5 before -
and the three regressions are bit-identical to p7w4. Gate 3's anchor
case is closed; bsl-esc-menu (M2) and the three-repeat final form remain.

docs/Disaggregated/CURRENT_STAGE_STATUS.md is new: the nine-item exit
denominator with evidence, what landed, what is running, what is next.
The ROADMAP's P7, Ph and P3b/P4b rows carry the wave-2 landing, the 6/9
checkpoint and the open items, and OQ-8/10/13 are marked answered.
2026-09-22 19:02:44 -04:00
swung0x48 dbea871a06 [docs] (P7): §12 - the one present per bsl-esc-menu replay is the trace's own two swaps and the harness's snapshot exit, so server reclaim must not assume a bounded frame 2026-09-22 18:56:11 -04:00
swung0x48 713bea9a8b [docs] (P7): ID-P7-33 - B3 landed: the OpenRA divergence was an unsound frame-serial floor, proven by counter and closed on both submissions 2026-09-22 18:54:26 -04:00
swung0x48 9a494b6f01 [docs] (P7): magma-b3 - the wire arm's frame-serial floor was raised on a fence it had not waited
Mechanism with file:line and the submission structure measured rather than assumed, the
confirmation tallies (zero unforced, which refutes the silent-draw-drop hypothesis the
investigation ranked first, and which the judge refuted independently on the prefix-tear
forensics), the provable floor, the fence-backed second term in WriteWireBuffer, the red-once in
three forms (a deterministic counter 26/26/0 -> 0/0/0, a two-process lane case, and forced OpenRA
retraces), the gate numbers, and the device acceptance with three predictions H1 must satisfy.

Records honestly: an earlier draft of this note blamed TryDrainFrameTransients' every-8th-drain
frame boundary, which this package's own probe refutes (one successful drain in the whole replay,
zero frame-boundary ones, all 33 serial bumps from Present); the GPU-drag knob the judge suggested
was not built, with the reason; the aggregate-wait secondary fix is left to a separate package
with the lane evidence that nothing in this tree's coverage regressed without it; and M1's
"present fires once" observation does not hold for OpenRA, whose pbuffer replay presents every
frame.
2026-09-22 18:44:07 -04:00
swung0x48 511c1f1641 [MG_Backend] (P7): the completed-frame-serial floor only advances to a serial nothing in flight still carries
THE OpenRA DEVICE DIVERGENCE. "Frame-serial completion piggybacks on submission completion"
holds only while a frame serial has ONE submission. On the wire arm it has two, measured on this
trace: 73 submissions over 32 serials on inproc and on spawn, 30 over 32 on monolith.

S1 is the per-frame palette glTexImage2D, which reaches UploadPendingWireLevels ->
FlushWirePendingCommandsForTextureUpdate -> FlushPendingCommands at the first palette-sampling
draw and submits under a POOLED fence. S2 is Present, and it carries the frame's draws AND the
barriered vkCmdCopyBuffer that StagedWireRangeCopy records for every streamed glBufferSubData.
RegisterSubmit stamps both with the serial current at submit time, so they share it.

S1 is tiny and signals almost immediately. RefreshCompletedSubmits retires it alone, and
OnSubmitsCompletedUpTo called NotifyFrameSerialComplete per retired record - advancing the floor
to N-1 while S2(N-1), the submission actually carrying frame N-1's buffer copies, was still
executing. GetCompletedSerial then returned N-1, so frame N's first write to that buffer failed
WriteWireBuffer's busy predicate and took the unordered host memcpy at :268 - and S2(N-1)'s
queued copies landed on top of it afterwards.

That is why the wrong pictures are a PREFIX TEAR of one draw's vertex range rather than a missing
draw: 318 vertices / 6 = 53 quads at 11448/53 = 216 bytes each, and 6696 = exactly 31 of them.
6696 and 11448 are both real write sizes of that same buffer (26 and 5 occurrences in this
replay), so the damage is the previous frame's write list replayed over the new bytes, and its
length is the largest write in whatever suffix of that list had not executed yet. It also
explains the device's FRAMESINFLIGHT=4/8 result (6/6 whole-write loss): more frames in flight
puts the GPU further behind, so the pending suffix contains the larger writes.

The floor now advances only to a serial NO remaining in-flight submission still carries.
NotifyFrameSerialComplete is monotone and already refuses the still-recording serial, so this can
only advance the floor later, never further; the disaggregated build's monolith arm is
behaviour-identical (one submission per serial, and its counter reads zero either way). The #else
branch is the pull build's statement verbatim, which is what keeps G1 byte-identical.

The counter is the red-once's reading, per lens C: an advance naming a serial some remaining
submission still carries. Base 26 on inproc, 26 on spawn, 0 on monolith; zero on all three after
this commit. It is logged per event rather than dumped at teardown because this trace's renderer
is never Shutdown() - the process exits - so a teardown dump reports the start-up context's zero.
2026-09-22 18:44:07 -04:00
swung0x48 a2e244c036 [MG_IntegrationTest] (P7): the streamed subdata-then-draw pair is a case on all three Magma arms
F1WireScenario.StreamedBufferSubDataBeforeEachDrawIsOrdered is OpenRA's frame reduced to four
quads: one GL_DYNAMIC_DRAW VBO that is never orphaned, four batches written over the SAME
offset-0 range and each drawn immediately, each in its own two-pixel column and its own colour.
The case asserts PIXELS, because this path has no GL error and no log line to assert - a draw
that read another batch's bytes still executes and still reports success.

Two entries per arm. The plain Fm. entry is a regression gate that runs on every driver. The
StaleSerial. entry carries MGITEST_MAGMA_FORCE_STALE_BUFFER_SERIAL=1 and is the rule-J red-once:
on integration-magma-spawn, at the package base, batches 0..2 lose their own vertices and batch 3
- the last write - keeps them, which is the mechanism spelled out in pixels. Its own ENVIRONMENT
rather than a lane-wide flip, for the reason the ShaderMip pair gives: the knob changes how every
streamed write in the process is answered.
2026-09-22 18:44:07 -04:00
swung0x48 c9164f25d3 [MG_Backend] (P7): a streamed buffer write waits for the submission that reads it, not for a frame count
WriteWireBuffer answers a resource_subdata either with an ordered vkCmdCopyBuffer or with an
immediate host memcpy, and it chose on `lastUseSerial > GetCompletedSerial()` alone.
GetCompletedSerial is max(m_frameSerial - frameCount, m_completedSerialFloor) and its first term
is PURE COUNTING - it asserts that work from frameCount serials ago has finished with no fence
behind it. On the wire arm m_frameSerial is advanced not only by Present but by
TryDrainFrameTransients -> VkBufferManager::BeginFrame on every 8th drain, which happens
MID-FRAME, and package B established that mid-frame pooled-fence submissions are not covered by
frame.imageInFlightFence. So the predicate could answer "idle" about a buffer that a
recorded-or-in-flight draw still reads; the write was then applied unsynchronised and the draw
executed reading somebody else's vertices - drawing another batch's geometry, or nothing at all,
with no log line and no GL error.

That is OpenRA's frame: ~24 glBufferSubData -> glDrawArrays pairs over the SAME range of one
GL_DYNAMIC_DRAW buffer that is never orphaned, and a picture missing exactly one batch's quads.

The predicate now has a second, fence-backed term: WireBufferResource carries lastUseSubmitIndex,
set at every draw bind and staged copy to GetWireNextSubmitIndex() (m_submitCounter + 1 - what
RetireWireObjects already tags future objects with, and deliberately NOT GetSyncPointSubmitIndex,
which answers m_submitCounter when nothing is recorded yet, exactly the hole a draw acquiring its
vertex slices before BeginCommandRecording falls into), and IsSubmitIndexComplete reports an
unsubmitted index as incomplete. The counting term is kept because it only ever over-reports
busy; the two are OR'd, so the answer is monotonically safer. GetCompletedSerial itself is
untouched - it also serves the monolith VkBufferResource path and changing it would move the pull
build's .text.

MGITEST_MAGMA_FORCE_STALE_BUFFER_SERIAL forces the counting term to answer "idle" (all writes, or
the Nth one alone), which is the state a mid-frame idle drain puts the device into and which no
host lane reaches by itself. It forces only the lie, not a bypass of the ordered copy, so it stays
a valid red-once across this fix.
2026-09-22 18:44:07 -04:00
swung0x48 76bbb6c300 [MG_Backend] (P7): every silent exit on the Magma wire draw path has a name, a tally and a decline line
Rule I admits two shapes for a refusal on the wire arm, a decline and a named Fatal. The draw
path had a third one the rule does not allow: roughly forty bare `return false` statements with
no log, no counter and no census entry, so a dropped draw left no trace anywhere - which is how
the OpenRA device divergence survived four device windows with client and server logs that were
line-identical between a passing and a failing run.

WireDeclines.def names every one of them; WireDeclineTally.h expands that into a counter table
plus MGL_WIRE_DECLINE_AT (the tally always moves, the log is once per site), in the shape of
PipeApplier.h's ServerVerbSink tallies and for the same reason (R-16: a probe may not arm
against a stub). Covered: SetupWireDraw and DispatchWireCompute (25), the three exits of
PrepareWireTextureResources, the four of SyncTextureResourceByHandle, and UploadPendingWireLevels'
five false exits PLUS the two that only `continue` - those report SUCCESS while leaving the entry
pending, i.e. a draw that samples a level which was never uploaded, so they are counted apart.

Also the pending-upload gauges (entries, bytes, peak bytes), sampled at the funnel, because a
refused upload does not consume its entry and the level shadow behind it stays alive.

Shutdown dumps the tally and is silent when it is all zero: a green lane stays quiet and a lane
that dropped a draw cannot. Plus the drain probe behind MOBILEGL_MAGMA_WIREBUF_PROBE, which is
what measured that a host replay reaches the mid-frame frame-boundary drain zero times.
2026-09-22 18:44:07 -04:00
swung0x48 b089e0de9b [docs] (P7): ID-P7-31 carries B2's review verdict and the rework it sent back 2026-09-22 18:43:20 -04:00
swung0x48 fe9105350c [docs] (P7): ID-P7-32 - bsl-esc-menu is a dead-VkBuffer leak on the wire arm, and the 40 % checkpoint passed at 6/9 2026-09-22 18:39:03 -04:00
swung0x48 85ac7e7692 [docs] (P7): ID-P7-31 - B2 landed: the six §3.2 retirements are complete, the ratchet re-baselines at 173, and the judge's Magma debts are recorded 2026-09-22 18:34:37 -04:00
swung0x48 116dc48e80 [docs] (P7): W4 §3b - the judge's verdict: an unsound completed-frame-serial floor lets a streamed write take the unordered host path
Four lenses and an adversarial judge: the two wrong OpenRA pictures are
prefix tears of one draw's vertex range, produced when frame N-1's first
pooled-fence submission retires and raises the buffer manager's floor
while the Present submission carrying the staged copies is still running,
so frame N's first glBufferSubData takes WriteWireBuffer's host-memcpy path
and the queued copies overwrite it. FIF=4/8 fail 6/6, as that predicts.
The later device sessions' logs (spawn, texture dumps, bsl memory, FIF)
join the evidence directory.
2026-09-22 18:30:46 -04:00
swung0x48 3c2867d34d [MG_Backend, scripts, docs] (P7): in a disaggregated build the monolith arm blits and mips with the baked modules, and the ratchet's p7-magma bucket falls to 88
CONTRACT-P7 §5.2's (B'), which §4.2 needs.

WHAT MOVES. In a disaggregated build the monolith arm now drives the SAME baked
modules the wire arm does:
  * InitializeBlitResources / InitializeDepthMipmapResources return true - there is
    nothing to build on any transport;
  * TryBlitToDefaultFramebufferWithShader turns its two BlitImageBindings into
    WireImages and calls BlitWireColorToDefault (WireColorBlit.inc);
  * GenerateDepthMipmapWithShader loops GenerateWireDepthMipLevel
    (WireDepthMipmap.inc) and hands the image back in the finalLayout its caller
    promised;
  * GetOrCreateBlitPipeline, the three GLSL strings, the eight kHidden* ids and the
    frontend members of BlitResources / DepthMipmapResources compile out.
The `#else` is the pull build's code, statement for statement.

WHY AN `#if` AND NOT A RUNTIME BRANCH - this is the whole of §4.2. P5f fv already
gave both initializers a non-monolith early return, and the ratchet counted the
symbols anyway, because a reference is a reference whether or not the branch runs.
And the MEMBER has to go with the construction: a SharedPtr member's destructor is
itself a reference to ~ProgramObject.

THE BUCKET TABLE (scripts/link_ratchet.py --bucket, build-split):

  bucket             before   after   a6
  p6-core                11      11    6
  p7-magma              101      88  101
  p3b-p4b-espryt         13      18   14
  both-backends          61      56   60
  total                 186     173  184

p3b-p4b-espryt's +5 is not a regression: those five MOVED out of both-backends
because DirectVulkan stopped referring to them and only DirectGLES still does.
Thirteen symbols went away outright, and they are exactly the three classes §4.2
names: ProgramObject::{AllocateLifetimeId, AttachShader, Link, ~ProgramObject},
SamplerObject::{SetLodRange, SetWrapR, SetWrapS, SetWrapT} and
ShaderObject::{Compile, DropCompileNode, JoinPendingCompile, ReleaseCompileNode,
SetShaderSource}. Re-baselined in this commit with --write-baseline; the 7 `# P13`
annotations carried over.

§4.2's "down to 0" DOES NOT HOLD, and the gap is not (B') left half-done - it is
that 88 of the 101 were never about the hidden programs. 62 come from
UniformManager.cpp.o and 14 more from VkSamplerManager.cpp.o: the MONOLITH
descriptor binder and sampler cache reading the application's own GL texture and
sampler objects (TextureObjectBase 35, TextureObjectWithOneMipmap 20 with its
vtable and typeinfo, SamplerObject's fourteen getters). Under a transport
SetupDraw branches to SetupWireDraw and not one of those lines runs - but that
branch is a RUNTIME one, which ID-P7-5 deliberately leaves to the optimisation
phase. notes/p7/magma-b2.md §4.3 names all eleven referring-object groups and what
each is.

NONE OF THEM IS ANNOTATED `# P13`, on purpose: that mark is for a symbol that
cannot fall without moving pull `.text`, and nothing above has shown that.
UniformManager's and VkSamplerManager's monolith arms are pull-build code, and
`#if`-ing them is the same operation this commit performed on VulkanRenderer.cpp.
They are wave 3's work, not P13's boundary problem, and marking them P13 early
would record an untried thing as a tried-and-impossible one.

Gates: integration-gpu 1705 (every monolith `DirectVulkan.` entry is in there),
integration-magma-split 93, -spawn 72, -tcp 74, integration-magma-full-split 524,
ctest -L unit 2423, all green; fatal_census 79 abort sites / 44 family words / 0
unmarked; spawn_lane_parity green; link_ratchet --assert-monotone unchanged at 173
against the re-baselined file, and --self-test OK.

RETRACE, which is this slice's load-bearing reading because the monolith arm
changed implementation: OpenRA.DirectVulkan (monolith) ssim 1.000000,
mismatchPixels 0, twice; OpenRA.DirectVulkan.SPLIT ssim 1.000000, 0. The whole
DirectVulkan matrix is 39/40 monolith and 38/39 split, and both reds are the SAME
case, minecraft-1.21.4-fabric-iris-iterationrp-in-world (ssim 0.027915), which is
PRE-EXISTING - it fails identically on ~/w7/p7-magma-b (this branch's parent),
because it needs CONTRACT-P7 §2.4's three iterationRP knobs and those are exported
by the CI job rather than by the ctest entry.

G1 pull .text 0xa52203, nm --defined-only identical to p7-before/pull-syms.txt
(30570 lines, 0 diffs).

Device debt: the monolith arm in the Android disaggregated package now runs the
baked blit, and its ROTATE_90/180/270 geometry has never executed - every surface
on this host is IDENTITY. notes/p7/magma-b2.md §5 lists it with the package's three
other device items.
2026-09-22 18:29:22 -04:00
swung0x48 a50d50df60 [MG_Backend, MG_Impl] (P7): the two Magma allocator debt scopes go, because the objects they were about are never built off the monolith arm
CONTRACT-P7 §5.2's "the two MagmaP7AllocatorDebtScope are deleted with it (prove
first, with a temporary assert, that they never arm under a non-monolith
transport)".

WHAT THE DEBT WAS. The hidden blit program, its two samplers and the hidden depth
-mipmap program are FRONTEND objects. Under an active transport the server backend
created them on the apply thread and destroyed them on it, where their destructors
run the client death helper and its lifetime-id probe - so P5c gave the two
teardowns a named exemption and P5e ruling 12 renamed it after the debt.

WHAT RETIRES IT IS NOT AN EXEMPTION BUT AN ABSENCE. P5f fv already gave
InitializeBlitResources and InitializeDepthMipmapResources an early return for
every non-monolith transport, so off the monolith arm neither resource ever holds
anything and `m_blitResources = {}` destroys nothing. The scopes have been
markers over a path that cannot run since that day.

PROVED, NOT ARGUED. A temporary probe in both teardowns failed the session by name
if either resource was non-null under a non-monolith transport. With it armed:
integration-magma-split 93, -spawn 72, -tcp 74, integration-magma-full-split 524,
-full-spawn 524, integration-split 187 and integration-magma-buffers 21 all green -
it never fired. The negative control is what makes that a measurement: dropping the
probe's transport test killed EVERY monolith `DirectVulkan.` entry at teardown
(8/8 aborted, after the case itself passed) while the split entries stayed green.
integration-magma-full-tcp's one red under the probe is the documented StreamLink x
Magma exposure texel (CONTRACT-P7 §2.5's named tcp entry,
IterationRPProgram203Scenario), not a probe death - it fails, it does not abort.

THE COUNT AGAINST P5e RULING 12. That ruling named FOUR Magma sites
(VulkanRenderer.cpp x3 and ResourceTracker.h x1). Two had already gone before P7
began - `git log -S` puts the ResourceTracker.h site's removal in P5f fv (4f3d2d6b)
and the third VulkanRenderer site's in P5f fm (036c3a3d) - so the P7 baseline
78b7d6be carried TWO, which are the two this commit deletes. Magma's apply-thread
allocator debt is now ZERO sites.

THE TYPE STAYS, and its comment now says why: MG_Test/Wire/RemoteClientTest's
RemoteGuards.BarrieredLegacyScopesCannotExemptAllocator and
.BarrieredFrontendRegistryMembersRefuseBothLegacyScopes construct it on the apply
thread and assert the allocator refuses ANYWAY. That is a negative control for the
property P5f (fr) established, and deleting the class would delete the only case
that pins it. No backend use remains.

Gates: integration-magma-split 93, -spawn 72, -tcp 74, integration-magma-full-split
524, ctest -L unit 2423, all green (one InitialCapsStartup death test flaked once
under -j 4 beside a sibling worktree's build and is green on re-run and on a second
full lane); fatal_census 79 abort sites / 44 family words / 0 unmarked;
link_ratchet --assert-monotone unchanged at 186; G1 pull .text 0xa52203, nm
--defined-only identical to p7-before/pull-syms.txt (30570 lines, 0 diffs).
2026-09-22 18:29:22 -04:00
swung0x48 51fa513d73 [MG_Backend, MG_IntegrationTest, MG_Test] (P7): a multisample depth or stencil buffer resolves on a device with no resolve extension, and the sample-count shapes with no arm decline
CONTRACT-P7 §3.2's `multisample-blit-aspect@P7`, the last `@P7` refusal in the tree.
Two halves again, and neither is the other's fallback.

THE HALF WITH AN ARM. vkCmdResolveImage is COLOUR ONLY. The one way Vulkan
resolves a depth or stencil aspect without a shader is a render pass with a
depth/stencil resolve attachment, which needs VK_KHR_create_renderpass2 and
VK_KHR_depth_stencil_resolve - and a device with neither ended the session at
`Magma:multisample-depth-resolve-capability` on the first
glBlitFramebuffer(GL_DEPTH_BUFFER_BIT) out of a multisampled read framebuffer,
which is what every multisample depth resolve in GL is. §3.2 names the second arm
and this is it: WireMultisampleResolve.inc, three baked modules (one vertex stage,
one fragment stage per aspect), fetching SAMPLE ZERO - the same sample
VK_RESOLVE_MODE_SAMPLE_ZERO_BIT picks, so the two arms answer identically and the
lane can assert that.

The pass carries nothing: the scratch has the source's extent and coordinate
space, so the fragment stage indexes by gl_FragCoord and the rectangle is the
scissor. No push constants, no varyings, no frontend object. The stencil half
needs VK_EXT_shader_stencil_export to exist at all (a fragment shader cannot write
the stencil aspect otherwise); where it is absent the arm declines stencil rather
than leaving the aspect undefined, and the module is not even created, because
handing vkCreateShaderModule SPIR-V with an unsupported capability is undefined
rather than an error.

The two capability tests are deliberately not the same test: the subpass resolves
every aspect the FORMAT has and needs a mode for each, while the pass writes only
the aspect this blit asked for - which is safe because the copy-out reads exactly
that aspect.

THE HALF WITH NO ARM. Sample counts that differ with the DESTINATION multisampled:
one sample fanned out to N, or N reduced to M. vkCmdResolveImage only ever writes
a single-sample destination, vkCmdBlitImage refuses a multisampled side, and
vkCmdCopyImage needs the counts equal. Rule I (a): name both counts, record
INVALID_OPERATION, leave the destination alone.

Lanes. DepthStencilReadbackMatrixScenario.AResolvedMultisampleDepth… is PROMOTED
from the informational tier into the three gated Magma arms - it already owns the
observable (clear a 4x DEPTH24_STENCIL8, resolve with glBlitFramebuffer, read both
aspects back, check that a depth-only resolve left the destination's stencil alone
and vice versa), it just had no gated entry. Two entries over it per arm: plain,
and under MGITEST_MAGMA_FORCE_SHADER_DEPTH_RESOLVE, which is the only way the new
arm is reachable here at all - lavapipe carries BOTH resolve extensions, so the
extension arm answers every resolve on this host. The new F1 case is the decline.
It does NOT skip on this host: the shader arm resolves depth and stencil, because
lavapipe also carries VK_EXT_shader_stencil_export.

Three rows join MOBILEGL_BAKED_INTERNAL_SHADERS (8 -> 11).

Red once (R-16, executed and reverted), three times:
  * restore MagmaWireFatal("multisample-depth-resolve-capability") for the
    non-extension arm -> DirectVulkan.Split.MsResolve1.… dies by that name while
    MsResolve0 stays green, which is also the proof that the knob routes;
  * restore MagmaWireFatal("multisample-blit-aspect@P7") ->
    DirectVulkan.Split.MsBlit.…FromASingleSampleSourceDeclines… dies by that name;
  * fetch sample 1 instead of 0 in WireMultisampleStencilResolve.frag (no rebuild -
    the gate reads the source at run time) -> WireMultisampleStencilResolveFragment
    reds at word 158, naming the symbol, the header and the regeneration command.
Reverted, 15/15 green on the three arms and 11/11 on the freshness gate.

`grep -rn "@P7" MobileGL/ --include=*.cpp --include=*.inc --include=*.h` now
returns COMMENTS ONLY - 19 lines, every one of them prose about a retirement or a
red-once. No refusal in the tree carries the mark any more.

Gates: integration-magma-split 93 (+3), -spawn 72 (+3), -tcp 74 (+3),
integration-magma-full-split 524 (+1), ctest -L unit 2423 (+3), all green;
bake_internal_shaders.py --check reports all five headers fresh; fatal_census 79
abort sites / 44 family words / 0 unmarked; link_ratchet --assert-monotone
unchanged at 186; G1 pull .text 0xa52203, nm --defined-only identical to
p7-before/pull-syms.txt (30570 lines, 0 diffs) - every line lives under
`#if MOBILEGL_BUILD_DISAGGREGATED` or in the test tree.

Device debt: the shader arm is the arm the PHONE takes wherever
VK_KHR_depth_stencil_resolve is missing, and this host can only reach it through
the knob. Its geometry needs one look on the Redmi.
2026-09-22 18:29:22 -04:00
swung0x48 452d33b578 [MG_Backend, MG_IntegrationTest] (P7): a scaled multisample colour blit resolves into a scratch and blits that, and a multisample destination declines
CONTRACT-P7 §3.2's `multisample-blit-shape@P7`. The name covered two halves that
are not the same question.

The half with an arm: a multisample colour blit whose destination rectangle is a
different size from its source. vkCmdResolveImage carries one extent and one
offset per side, so it can express neither the scale nor the flip - which is why
this arm already resolved into a single-sample scratch at raw offsets and blitted
THAT for the flip. The two `width != |dx1-dx0|` clauses were the only thing
keeping the scale out of the same route, and the monolith arm reaches it the same
way (VulkanRenderer.cpp's `regionWasTransformed` branch). They are gone; the
scratch is still sized by the SOURCE rect, so only the second leg sees the scale.

The filter only started mattering with that scale - at 1:1 LINEAR and NEAREST read
the same texel - so the second leg now asks whether the format can be filtered
before it passes VK_FILTER_LINEAR, which is undefined behaviour on a format whose
optimalTilingFeatures lack SAMPLED_IMAGE_FILTER_LINEAR. GL ignores filter for a
multisample resolve anyway, so the fallback is NEAREST and not a refusal.

The half with no arm: a MULTISAMPLE DESTINATION. vkCmdCopyImage is the only
command that writes one, it takes a single extent, and GL 4.6 core 18.3.1 makes a
blit whose rectangles differ in size INVALID_OPERATION when either framebuffer is
multisampled. There is no picture to get right, so rule I (a): MGLOG_E_ONCE with
both rectangles and both sample counts, the error GL names, the destination left
alone, the session kept.

Two entries on each of the three Magma arms (`MsBlit.`), split-only because the
monolith arm drops the scale (its vkCmdResolveImage lands 1:1 at the destination
offset), so an equality between the arms would assert that the wire arm is as
wrong. The scale and the flip share one case over one source: dropping the scale
lands an 8x4 band in a 4x2 attachment, dropping the flip returns the right two
colours in the wrong order, and the second blit's expectation is the first's with
the rows swapped.

Red once (R-16, executed and reverted), twice:
  * restore the two size-equality clauses ->
    DirectVulkan.Split.MsBlit.…ScalesAndFlipsThroughTheResolveScratch dies as
    Fatal{UnmigratedVerb, "Magma:multisample-resolve-region"};
  * restore MagmaWireFatal("multisample-blit-shape@P7") ->
    DirectVulkan.Split.MsBlit.…DeclinesAndKeepsTheSession dies by that name.
Reverted, 6/6 green on the three arms.

Measured on the way in: the decline's GL error reaches the client with a LATER
reply, not with the blit call, so the case's glFinish is load-bearing - without it
the INVALID_OPERATION surfaced attached to the 1:1 resolve twenty lines down.

Gates: integration-magma-split 90 (+2), -spawn 69 (+2), -tcp 71 (+2),
integration-magma-full-split 523 (+2), ctest -L unit 2420, all green;
fatal_census 79 abort sites / 44 family words / 0 unmarked; link_ratchet
--assert-monotone unchanged at 186; G1 pull .text 0xa52203 with an nm
--defined-only list identical to p7-before/pull-syms.txt (30570 lines, 0 diffs) -
every line of this commit is inside WireFramebuffer.inc's
`#if MOBILEGL_BUILD_DISAGGREGATED` or in the test tree.
2026-09-22 18:29:22 -04:00
swung0x48 22761a692d [MG_Pipe, MG_Backend, MG_Remote, docs] (P7): ID-P7-30 - D3's two review fixes: the bit-8 and bit-13 rows are pinned, and deletion ordering is guaranteed by the Live/Gen test
The vertex-input and buffer-bindings rows of SubsystemDeps.def became live
behaviour when every reader started reading the table; two static_asserts
now pin them the way the texture and programs rows already were. The
"safe by construction" story about destroy ordering was inverted for the
case where a view holds its owner's last reference - the owner's destroy
reaches the wire first - so the three places that told it now say that the
walk's Live/Gen test is the guarantee.
2026-09-22 18:27:43 -04:00
swung0x48 498eb66a14 [docs] (P7): ID-P7-29 - D2's rework landed on top of D3 2026-09-22 18:21:52 -04:00
swung0x48 a2d7fbd7ca [MG_IntegrationTest, docs] (P7): the census note carries its own gate readings and the right counts
Rework round 4 of 4 after the D2 review (fable, ID-P7-8).

COUNTS. The note said 71 cases per arm where its own table sums to 72, and "+37 entries
per split arm" - a number that matched nothing. The +37 came from differencing against
~/w7/pipe/build-split, the integrator's LIVE build directory, which had already moved past
this package's base (7ed5da52 -> 8c5dd6fc). Measured against the pinned
~/w7/p7-before/ctest-names-split.txt baseline this package adds 76 DirectGLES.{Split,Spawn,Tcp}.
entries per arm: 72 census + 3 TextureViewAlias + 1 TextureUploadShape. The raw diff shows
77; the extra one is CtWireScenario's kCapResidentSubData case from b56d70177, one of the
five P7 commits between that baseline and this package's base, and no CtWire file is
touched here. Both numbers and their provenance are now in the note.

GATES. The note pointed at espryt-d2.md for readings espryt-d2.md did not carry. It now
has its own table the way espryt-d1.md does: four lane totals, G1 .text and nm, the G14
name delta, and the five CI gates.

CMAKE. Drops the MGL_SPLIT_D2_FILTER_<Scenario> indirection, which was copied from the d1
and i1 blocks but never defined for any scenario here - a branch no run takes. Every D2
scenario is registered whole because the census found no case in any of them that needs
excluding; the d1/i1 shape is what to reach for the day one does.

And records that the TcpDevice arm gets the texture-view knob on its CLIENT half only. The
loopback Tcp arm's server is started by the TcpServer.Start fixture, which this package
also gives the knob; the TcpDevice arm's server is a supervisor already running on a
device, started outside this build, whose environment nothing here can reach. If it was
started without the knob every texture-view case in that arm green-skips - exactly what
the loopback arm did before the fixture entry got it. The arm is unconfigured in this repo
(empty endpoint, so it is not created at all), so this is a note for whoever configures it.
2026-09-22 18:17:54 -04:00
swung0x48 394df89237 [scripts, ci] (P7): the memo purity gate asserts its own anchors and allow-list, and stops calling a policy exemption an absence
Rework round 3 of 4 after the D2 review (fable, ID-P7-8). Three defects, each of a kind
the gate could not have reported about itself.

(a) TWO STRUCTURAL SELF-CHECKS. family_spans() returned whatever matched and nothing ever
asserted that each of the five FAMILIES anchors found its declaration, so a renamed struct
would drop a whole family while every rule kept passing. And an allow-list entry no rule
can consult documents a carve-out that is not happening - a claim about the tree that
nothing verifies. main() now fails on either, naming the anchor or the entry and saying
which of the two repairs is wanted. Both are exercised: renaming an anchor and adding a
dead entry each red the gate.

(b) THE DOCSTRING AND THE test.yml COMMENT WERE WRONG about what the guards mean.
MOBILEGL_PIPE_LEGACY_MEMOS defaults ON (CMakeLists.txt:43) and no job here clears it, so
CI split builds DO compile those arms - TwinLookupMemo (DirectGLES.cpp:114) sits inside
one and is compiled into every lane this repo runs. The rule is "the legacy arm MAY KEEP
pointer keys", a policy exemption; only !MOBILEGL_BUILD_DISAGGREGATED is code a split
build genuinely never sees. Both now say so.

(c) FRONTEND HAD NO LEADING WORD BOUNDARY, so BackendSamplerObject* matched SamplerObject
and the gate reported the memos' backend VALUES as frontend keys - which the first version
answered with two allow-list entries, i.e. it documented a false positive instead of
fixing it. (?<!\w) added, with controls both ways: the backend twins must stay green and
the qualified MG_State::GLState::ProgramObject* must still match.

The allow list therefore goes from six entries to ONE. Four were dead on arrival - two
false positives now fixed at the source, one naming a typedef with no frontend type in it,
and const void* program / Uint64 programLifetimeId, which no rule could ever consult
because void is not in FRONTEND and neither line calls GetLifetimeId(). The
ResolvedTextureBindingMemo monolith key those last two described is still deliberate and
still there; it simply never needed an exemption, and saying it did was misleading. What
remains is the twin registry's UnorderedMap<StateObject*, Entry>, which really is
suppressed and really is the pull build's only arm.
2026-09-22 18:17:25 -04:00
swung0x48 0f3b82725d [MG_IntegrationTest, docs] (P7): the upload-shape gate was measuring its own bounding box; inset the scatter so the box/rect decision is reachable
Rework round 2 of 4 after the D2 review (fable, ID-P7-8), which found that this gate's
numbers were right and its SUBJECT was wrong.

The scatter wrote 40 2x2 rects at x=((i*7)%32)*2, y=((i*5)%32)*2 into a 64-px atlas.
Both strides are surjective onto 0..31 over 40 iterations - rect 9 lands at x=62, rect 19
at y=62 - so the rects' union box covered the WHOLE LEVEL, and the server's subRectEligible
(Managers.cpp:8486-8491) requires !dirtyRegion.CoversWholeLevel, a BOUNDING-BOX test
(MipmapStorage.h:30-33). Both scatter textures were rect-INELIGIBLE on every arm before
:8649's unpack-ring policy was ever consulted. The box-versus-rect decision the scenario
exists to pin was never reached for the two textures it scatters into, and the only
texture that could move under any red-once was the contiguous control band - which is
exactly what the old substitute red-once's box=6 rect=3 jobs=12 was.

kScatterMargin insets the scatter by one rect: same 40 writes, same coarse stride, union
box bounded at (2,2)..(62,54). The gold row is UNCHANGED at emit=9 box=9 rect=0 jobs=9 on
all four arms - but it now means what it says.

THE RED-ONCE IS NOW ZERO-PATCH, which is stronger than the throwaway source edit it
replaces: MOBILEGL_ESPRYT_DISABLE_UNPACK_RING=1 alone gives box=3 rect=6 jobs=183 with
bytes/f[tex] falling from 104448 to 9024, while all 26 pixel cases stay green. BOTH
SCATTER TEXTURES moved - box=3 is the contiguous band alone, rect=6 is the plain atlas
AND the atlas written through the glTextureView, and jobs=183 = 3 + 6x30 confirms thirty
rects per scatter emission. The same knob was green before the inset, which is the
evidence that CoversWholeLevel was short-circuiting the decision.

The three prose sites are rewritten accordingly: the scenario's gold-table rationale, the
MEASUREMENTS.md 13.1 reading and the package note's red-once section now explain
attempts 1-3 as mutations of a decision the client already makes correctly, and attempt 4
by CoversWholeLevel rather than by the ring.

Also PipeStatsWindow gains LastFrom{Client,Server}LogSince, and the gate reads its two
role windows since a mark taken before the counted workload. The server writes an EMPTY
teardown window after the counted one, so "the last line in the file" is one shutdown
away from reporting four zeroes as the workload's shape. Moving to a marked read exposed
a second latent defect in the same assertion: ASSERT_TRUE(clientWindow.found) over the
whole file was satisfied by the "counters ON" BANNER in all three split arms, claiming a
window had been read when none had. The assertion is now on the window that actually
carries the tex[] bracket.
2026-09-22 18:17:25 -04:00
swung0x48 b3334cf336 [MG_Test] (P7): SubsystemDepsTest skips its two resolver cases in the pull build, and says what keeps the sweep alive
Rework round 1 of 4 after the D2 review (fable, ID-P7-8). Mirrors the integrator's
6da19f71 so that later commits from this worktree apply cleanly, and takes the two
review notes attached to the same file.

THE GUARD. The four Resolve<Family>SubsystemArm() declarations live inside
did not compile in the pull build (ID-P7-11). Both now GTEST_SKIP under
!MOBILEGL_PIPE_PUSH the way the client-side case already did. Verified by re-running
the file's own compile command with -DMOBILEGL_PIPE_PUSH=0 and MOBILEGL_BUILD_DISAGGREGATED
undefined: -fsyntax-only rc 0.

WHAT KEPT THE MASK SWEEP ALIVE, now asserted instead of assumed: every resolver ends in
ClassifyPipeSubsystemArm, and a NoArm verdict calls StopOnArmlessPipeSubsystem
(Managers.cpp:2928-2938), which ABORTS. The sweep deliberately drives masks that refuse a
family's handle arm, so it survives only because MOBILEGL_PIPE_LEGACY_MEMOS defaults ON
(Config.h:399) and leaves a legacy arm standing. ASSERT_TRUE on that, with the reason -
the day the default flips, this file would have aborted with no explanation.

THE FRAMEBUFFER RESOLVER IS RECORDED, NOT ASSERTED, and the asymmetry is the finding:
ResolveFramebufferSubsystemArm folds D-C3's wire-caps refusal in beside the dependency
rule (Managers.cpp:4323-4325), and that half asks the server's published caps, which a
bare unit binary with no session cannot answer in the affirmative. So a false says nothing
about D-K2 and asserting true would pin whatever the binary happens to have linked. The
sweep still covers it in the direction that matters. Also adds the 0x5ff refusal lane
beside 0x7ff.
2026-09-22 18:17:25 -04:00
swung0x48 310d4e90d1 [docs] (P7): ID-P7-28 - package D3 landed; the texture-view exclusion is retired 2026-09-22 18:16:55 -04:00
swung0x48 00d0239696 [docs] (P7): package D3's note, and the census row its fix falsified
P3b/P4b wave 2-D package D3, slice 3.

docs/Disaggregated/notes/p34b/espryt-d3.md, in espryt-d1.md's shape: the mechanism end to end,
the decision between resolving the predicate and indexing the reverse edge with the argument
that decided it (deletion ordering), the red-once evidence with the exact ctest invocations and
the red and green lines per arm, the gate numbers, and six things that remain.

The decision, stated once here because the note argues it at length: the gate resolves through
Desc.ViewOf to the storage record; the reverse index was rejected because it needs pruning on
VIEW death or an owner upload bumps a recycled slot's Serial, because it makes
ApplyTextureUpload write a record other than the one the call names, and because
VkTextureManager::ResolveWireTextureStorage already resolves the same way on the Magma side -
an index would have given the two backends two answers to one question.

notes/p34b/split-scenario-census.md is updated where the fix falsified it - the summary row,
the texture-family row and the exclusion section - and NOT rewritten anywhere else. D2's text
stands and D3's resolution sits above it in a quoted block, because a package note is a record
of what that package did rather than a live statement of the tree. Two citation corrections
taken while reading the device-window baseline: the CTS block is KHR-GL46 and not KHR-GL43, and
texture_view is 1/7 fail there with the failing case view_sampling, NOT coherency - coherency
reads Pass, which is what D2 claimed and is now checked per case against report-texture.json
rather than against a summary line.

espryt-d2.md itself is left alone and the note says why it is left alone.

Gates on this head, all measured rather than carried over: G1 .text 0xa52203, nm --defined-only
30570 symbols 0 added / 0 removed, -D 0/0, ninja -C build-linux rc 0. fatal_census rc 0 (79
abort sites, 0 unmarked - unchanged, rule I). link_ratchet --assert-monotone rc 0 (unchanged at
186). spawn_lane_parity rc 0. Lanes: unit 2418/2418, integration-split 264/264,
integration-spawn 180/180, integration-tcp 183/183, integration-magma-split 75/75, all 100%.
Name sets against a real ctest -N at a7c06f4f: +1 on each of split/spawn/tcp (the un-excluded
case), 0 elsewhere, 0 removed anywhere.
2026-09-22 18:10:56 -04:00
swung0x48 1c36987e84 [MG_Backend, MG_Pipe, MG_Impl] (P7): D-K2's dependency rule has one statement, and every reader reads it
P3b/P4b wave 2-D package D3, slice 2 - D2's two loose ends, now in-partition because D1 has
landed on feat/disaggregated and MG_Backend/DirectGLES/ is free.

THE STALE PROSE, in the header of the file that implements the refusal.
MG_Backend/DirectGLES/Managers.h's resolver block said "THREE OF THEM CARRY A DEPENDENCY"
(there are six rows; bits 8 and 13 were the two it did not know about) and listed
"10 without 11" among the mirror pairs that are FINE - when 10-without-11 is precisely D-K2's
fourth row (ID-14/ID-15), refused twenty lines below by ResolveTextureResourceSubsystemArm and
withheld by the client. MGPipe.h lost the identical pair of defects with R-5; this copy was
left behind because the file belonged to a package in flight. It does not restate the rows
again: the file that states them is the file the code reads.

BOTH READERS NOW READ MG_Pipe/SubsystemDeps.def. More than the two one-line changes D2
estimated, because the rule had six statements and the estimate counted two:

  PipeFill.cpp P4aFamilyDependencyBits   its own four-row kMGPipeP4aFamilyDependencies[]
                                         -> MGPipeSubsystemRequires(); table deleted
  PipeFill.cpp P5eFamilyIsLive           hand-coded (pushMask & bit 7) == 0
                                         -> MGPipeSubsystemDependenciesAreSet(bit 13, mask)
  Managers.cpp PipeSubsystemDependencyMissing   took a literal dependency BIT per call site
                                         -> takes the FAMILY and asks the table
  Managers.cpp the four resolvers        five calls, five hand-written sentences
                                         -> four calls, one per family, naming only the family
  Managers.cpp ResolveVertexInputSubsystemArm    hand-rolled bitSet && !resourcesBitSet
                                         -> the table
  DirectGLES.cpp ResolveBufferBindingSubsystemArm  the same hand-rolled pair -> the table

MGPipe.h gains one accessor, MGPipeSubsystemDependencyWhy(), so a reader that refuses prints
the ROW'S OWN sentence instead of a copy of it. The .def has carried a Why column since R-5
and nothing read it; its "WHO READS IT" section is rewritten because "THEY ARE NOT SWITCHED
OVER IN THIS COMMIT" is no longer true of anything.

THE CLIENT'S ARGUMENT IS MASKED TO kMGPipeP4aFamilySubsystems, and that mask is the whole
difference between a rewrite and a behaviour change. The .def carries SIX rows; the table
deleted from PipeFill.cpp carried FOUR, and that predicate is wants()'s fourth conjunct for
P4a's families only. Reading all six unmasked makes the CLIENT withhold bit 8's vertex-input
emissions at a mask with bit 7 clear, which is P3a's own rule and not this gate's - caught,
not reasoned about, by TextureEmitTest's
EveryDKTwoDependencyRowGatesItsOwnFamilyAndTheMirrorPairsStayLive going red on exactly that
assertion ("bit 8 alone, with bit 7 clear: P3a's own rule, which this table must not touch").
Bit 13 is not narrowed there either; it asks the table for itself in P5eFamilyIsLive, because
bit 13's conjunct is bit 13's own. The ROWS come from one place; WHO each gate speaks for is
still each gate's.

ONE BEHAVIOUR IMPROVEMENT FALLS OUT rather than being designed in: the texture family is the
only row with two required bits and used to be two guarded calls, so a mask missing bit 11 but
carrying bit 7 was reported with the bit-7 sentence. One call now computes required & ~mask
from the row, so the refusal names whichever bit is actually missing.

WHAT THIS COSTS, said out loud. SubsystemDepsTest used to compare two COPIES of the rule; it
now compares two READERS of one table and cannot catch a typo in a row. It is kept and the
.def says why: both sides still fold in reasons of their own - the server's own family bit and
D-C3's wire-caps refusal, the client's consumer and wired conjuncts - and those can still
disagree at a mask. The compile-time half of D2's red-once (drop the fourth row from the .def
-> MGPipe.h:225/:230 fail) is untouched and is now the load-bearing one.

SubsystemDepsTest 5/5, TextureEmit's D-K2 case green, unit lane 2418/2418.
G1: .text 0xa52203, nm --defined-only 30570 symbols, 0 added / 0 removed, -D 0/0. Every line
is inside #if MOBILEGL_PIPE_PUSH or #if MOBILEGL_BUILD_DISAGGREGATED, and MGPipe.h's new
accessor is inline constexpr and unreferenced in the pull build.
2026-09-22 18:10:56 -04:00
swung0x48 1e1f4bad27 [MG_Backend, MG_IntegrationTest, MG_Remote, docs] (P7): an owner-side upload shows through a sampled texture view, because the split clean gate resolves to the storage record
P3b/P4b wave 2-D package D3, slice 1. The defect package D2 found with its new
TextureViewAliasScenario gate and excluded by name because the fix is outside D2's footprint.

THE HOLE. Once a glTextureView had been SAMPLED, a later glTexSubImage through the OWNER's
own name never became visible through the view, on inproc, spawn and tcp, green under
monolith. The client half was never at fault: the case's own emit-counter assertions pass on
the red run, so the records were emitted and the applier accepted them. A view owns no texels
and the client keeps ONE emission cursor per storage - an upload through a view's name is
remapped onto its storage owner before it is emitted (MG_Impl/Pipe/TextureEmit.h) - so every
resource_subdata for either name is keyed on the OWNER and ApplyTextureUpload moves the
OWNER's Serial and PendingUploads alone. BackendTextureObject::IsDrawSyncCleanByRecord read
the VIEW's Serial and PendingUploads, which nothing an upload does can move, so the gate
answered CLEAN for every owner write after the view's first sample and the sync that would
have drained the owner's pending uploads never ran.

THE FIX, AND WHY THIS ONE. The gate's two STORAGE clauses are asked of the STORAGE record -
this record for a texture that owns its texels, the record Desc.ViewOf names for a view -
through the new PipeTextureStorageRecordForRecord. The three PARAMETER clauses stay the
view's own record's, because a view has its own texture parameters and its own built-in
sampler, which is the whole reason the second name exists. SyncTextureViewToBackendByRecord
stamps that same storage serial at both its arms, so the stamp and the gate name one
quantity. This is the split equivalent of the monolith gate's GetContentVersion(), which
TextureObjectView forwards to its owner (TextureObjectView.cpp:100-102).

THE ALTERNATIVE WAS REJECTED. A reverse index owner->views, so an owner-side resource_subdata
bumps every viewing record's Serial, is the other shape D2 named. It is wrong here on three
counts: it puts new mutable state in MG_Pipe's applier, which both backends read, to answer a
question one backend's gate asks; it makes ApplyTextureUpload write a record other than the
one the call names, which breaks the applier's own serial discipline; and its index has to be
pruned on view death or an owner upload bumps a recycled slot's Serial - a silent
cross-texture hazard with no test that would catch it. The resolve direction needs no state,
no pruning, and is checked at every use: PipeTextureRecordForHandle tests Live and Gen.

DELETION ORDERING, WHICH DECIDED IT. A view holds a STRONG SharedPtr to its storage owner
(TextureObjectView::m_storageOwner) and the client emits resource_destroy from the DESTRUCTOR,
not from glDeleteTextures - "the last SharedPtr to this object dropping, not the glDelete*
that only marks the name and leaves a still-bound object very much alive"
(TextureObject.cpp:63). So an owner's record outlives every record that views it whatever
order the application deletes the two names in, and the forward edge cannot dangle. The walk
checks anyway, and a null answer is NOT clean - the direction that re-syncs.

VIEWS OF VIEWS. ONE hop always reaches storage: glTextureView composes a view-of-a-view onto
the ROOT at creation (the spec's additive min-level rule), the client's emitter says so and
TextureObjectView's invariant is that its owner is never a view. The walk is transitive
anyway and BOUNDED at four hops, because this side may not depend on a client invariant to
terminate; a cyclic chain the wire could forge ends as a null answer, which is "not clean",
not a refusal. No new abort site and no new Fatal (rule I): fatal_census abort_sites stays 79.

RED-ONCE (rule J), the case D2 excluded, with the exclusion lifted and nothing else changed,
at D2's head a7c06f4f:

    ctest -R 'DirectGLES\.(Split|Spawn|Tcp)\.TextureViewAliasScenario' -j 3
    ->  79% tests passed, 3 tests failed out of 14
        4042 - DirectGLES.Split.TextureViewAliasScenario.TheOwnerUploadDrainRunsAgainAfterADrawBoundary (Failed)
        4046 - DirectGLES.Spawn.TextureViewAliasScenario.TheOwnerUploadDrainRunsAgainAfterADrawBoundary (Failed)
        4050 - DirectGLES.Tcp.TextureViewAliasScenario.TheOwnerUploadDrainRunsAgainAfterADrawBoundary (Failed)

    the first owner write, through the view: 11408 of 11408 pixels disagree;
      first at (2, 2) is rgba(50,160,30,255), expected rgba(0,0,255,255) +/- 2
    the second owner write, through the view: 11408 of 11408 pixels disagree;
      first at (2, 2) is rgba(50,160,30,255), expected rgba(255,255,0,255) +/- 2

Only the PIXEL channel is red; the drain's emit counters pass, which is what places the
defect on the server. The monolith control DirectGLES.TextureViewAliasScenario.* is 4/4 in
the same run. With the fix: 14/14, 100%.

CONTRACT-P5E.md 5.2's clean gate is amended to state the rule the code now implements, and
says out loud that the rec->Serial form it carried through P5e was wrong rather than
restating it. The CMake exclusion is lifted; the scenario registers all four cases on all
three arms again (G14: the name never left any lane).

G1: .text 0xa52203, nm --defined-only 30570 symbols, 0 added / 0 removed, -D 0/0. Every line
touched in Managers.cpp is inside #if MOBILEGL_PIPE_PUSH (the gate and the resolver) or
#if MOBILEGL_BUILD_DISAGGREGATED (the view sync), and Managers.h's declaration is inside the
PIPE_PUSH block, so the pull build's preprocessed text is unchanged.
2026-09-22 18:10:56 -04:00
swung0x48 550f81d5f8 [docs] (P7): W4 §3b - the OpenRA hole is one draw (call 30376) reading vertex bytes other than call 30366's
Lens D placed 100 % of the wrong pixels inside the screen-space union of a
single glDrawArrays whose vertex data was the frame's first glBufferSubData
into the streaming buffer; the palette respecify is not the mechanism. The
candidates are WriteWireBuffer's serial-gated branch and the mid-frame drain
that bumps the frame serial without a fence.
2026-09-22 18:00:19 -04:00
swung0x48 ebd73da06c [docs] (P7): the W4 evidence files are plain files, not executables
Copied off a Windows mount, they came in as 100755; nothing in them runs.
2026-09-22 17:57:57 -04:00
swung0x48 770f1a0b3a [docs] (P7): ID-P7-26/27 - D2 lands with a rework round, and bsl-esc-menu is a memory defect, not an exclusion
D2's fable review says land with fixes: the TUS gate's scatter workload has a
union box that covers the whole level (rect=0 is geometry, not the unpack-ring
policy), the memo purity script fails open on a renamed family, and the census
numbers were not reproducible - all sent back to D2 as one rework round; the
excluded TextureViewAlias case gets its §12 row (owner D3).

bsl-esc-menu-854 on the spawn arm: monolith passes at 807 MiB peak RSS, the
server process dies at ~1.2 GiB in scudo's secondary allocator with 8 GB of
system memory free, inproc dies at 1.6 GiB. That is the wire arm holding about
twice the bytes (or the mappings) of the monolith, a disaggregation defect
under ID-P7-12's rule, so it stays in the gate-3 denominator; M1 attributes it.
2026-09-22 17:57:22 -04:00
swung0x48 dd6c541af7 [docs] (P7): ID-P7-25 - W4 on p7w4: 34/36 match monolith, and OpenRA is a silent draw drop, not a cold cache
Device window 1 W4 (APK p7w4 = 8c5dd6fc): iterationt x2 pass, the 39-case
inproc matrix has 34/36 of the denominator within 0.0005 of the monolith
reading, bsl-esc-menu still OOMs under inproc (spawn arm pending).

ID-P7-19's cold-driver-cache reading of OpenRA is withdrawn: 26 runs show a
~50 % per-run coin flip that ignores cache temperature, IPC_AUDIT and FIF=1,
reproduces on the two-process spawn arm, and only ever yields three pictures
(golden / 9836 px / 14658 px). The monolith snapshot at the previous swap is
the golden and the palette texture is byte-correct at the snapshot, so the
wrong pictures are frames with a bounded run of draws silently skipped (nine
sprite blobs absent over intact terrain). Gate 3's acceptance for OpenRA is
now six run-ahead and six lockstep runs all equal to the golden; the fix is
package B3.
2026-09-22 17:51:56 -04:00
swung0x48 c2cc3a9673 [MG_Test] (P7): the subsystem-deps server-gate cases compile in the pull build too
D2's SubsystemDepsTest called the four Resolve*SubsystemArm resolvers, which
Managers.h declares only under MOBILEGL_PIPE_PUSH (:664), so build-linux
(the pull flavour, ID-P7-11) stopped compiling while the stale libMobileGL.so
kept G1 looking green. Cases 3 and 4 now skip in the pull flavour the way
case 2 already did; the unit name set is the same in both flavours. Pull
build rc 0, .text 0xa52203, symbols 0 added / 0 removed; push flavour 5/5.
2026-09-22 17:37:59 -04:00
swung0x48 519445f946 [MG_Pipe, MG_Impl, MG_Test, docs] (P7): D-K2's dependency rule in one table, with the case that fails when the two readers disagree
P3b/P4b R-5 (and R-3's cheap half), landing under the P7 program (PLAN-PH-P34B-P7.md
1.2, wave 2-D package D2).

The rule had SIX statements and nothing compared them, and two of the four prose ones
were wrong in the same way: MG_Pipe/MGPipe.h and MG_Backend/DirectGLES/Managers.h both
said "THREE OF THEM HAVE A DEPENDENCY" (there are six rows) and both said "the mirror
pairs (10 without 11, ...) are all fine" - when 10-without-11 is precisely D-K2's FOURTH
row (ID-14/ID-15), refused by the server and withheld by the client, with PipeFill.cpp
saying so out loud: "The brief's original 'bit 10 without 11 is fine' is WITHDRAWN for
P4a as built." Neither wrong comment could fail anything; Managers.h's copy sits in the
header of the file that implements the refusal.

The rows, once, in MG_Pipe/SubsystemDeps.def (X-macro, FatalFamilies.def's convention),
expanded by MGPipe.h into kMGPipeSubsystemDependencies[] / MGPipeSubsystemRequires() /
MGPipeSubsystemDependenciesAreSet() with four static_asserts. Six rows: 8->7, 9->10,
10->{7,11}, 11->10, 12->none (a ROW, not an absence), 13->7. Bits 8 and 13 are the two
the old comment did not know about; bit 13's row lived fifteen lines further down in the
same header, which is how "THREE OF THEM" survived two phases.

THE TWO READERS DO NOT YET READ IT, deliberately: MG_Impl/Pipe/PipeFill.cpp is the
contract package's file for the whole phase (TextureEmit.h states that rule) and
MG_Backend/DirectGLES/Managers.cpp belongs to wave 2-D package D1 while it is in flight.
Switching them is two one-line changes for the integrator. The anti-drift property does
not wait: MG_Test/Backend/DirectGLES/SubsystemDepsTest.cpp is the first thing in the tree
that calls BOTH readers at the same mask - sweeping the phase constants, the two
documented refusal lanes and one mask per row - and compares each against the .def. It
lives in that binary because that binary links both sides.

Both comparisons are one-directional, each for a measured reason. The server resolvers
also fold in the family's own bit and D-C3's wire-caps refusal, so a resolver may say NO
for its own reasons but never YES where the table refuses. MGPipeP4aFamilyEmits is a
QUADRUPLE and the first version of this case asserted equality and reddened at 0x1fff on
the sampler and program families because a bare unit binary registers no consumer - the
assertion being wrong, not the product, and recorded in the note because the temptation
to assert equality here is strong.

Red-once (R-16), twice: dropping the fourth row from the .def fails at COMPILE time
(MGPipe.h's static_asserts); making the server's ResolveSamplerSubsystemArm stop checking
bit 10 (throwaway, reverted) reds the runtime case and names the mask - 0x9ff, which is
exactly the lane CMakeLists.txt already pins for G12.

R-3's cheap half: CONTRACT-P5.md claimed the image-access, sampler-view-target and
draw-buffer encodings as contract by pointing at package-header LINE RANGES, and all
three had drifted (ImageEmit.h:146-159 no longer holds the encoding at all). All three
rows now state the encoding by enumerator and function name, and the two confusable
siblings are named. The remaining half - MGPSamplerView::Target still has no Count bound,
no ...IsValid and no Fatal for an out-of-range value - is stated in the row rather than
closed, being an MG_Pipe change outside this package's footprint.

Managers.h:690-695's copy of the wrong prose is NOT corrected here: it is package D1's
file, and its two defects are named in notes/p34b/espryt-d2.md so the integrator can take
them in one line.

G1 after this change: pull .text 10822147 = 0xa52203, nm --defined-only identical to
~/w7/p7-before/pull-syms.txt (30570 symbols) - the new inline constexpr table does not
reach the pull build.
2026-09-22 17:31:51 -04:00
swung0x48 779ff21b3e [MG_Test] (P7): the raw-depth-fetch sampler case runs both arms; the deletion half is P13
P3b/P4b deepening, landing under the P7 program (PLAN-PH-P34B-P7.md 1.2, "raw-depth-fetch
sampler 的 monolith 清理,步骤 1+2", wave 2-D package D2).

STEP 1 DONE. ContextEpochTest gains a case that drives GetRawDepthFetchSampler() on BOTH
arms in one process and asserts they are two different BackendSamplerObjects, which is the
observable form of the claim DirectGLES.cpp:334-336 makes in prose - "the legacy pair is
not constructed or consulted by the server". A server that had fallen through to
g_rawDepthFetchSamplerBackend would hand back the same pointer under both transports. It
also asserts the transport arm's cached native sampler survives a round trip through the
monolith arm, so neither arm is rebuilding state the other owns.

STEP 2 NOT DONE, AND THE PACKAGE'S OWN CONDITION IS WHY. The instruction was to delete what
the transport arm makes unreachable "ONLY if it keeps the pull build's .text identical".
It does not. Every candidate is UNCONDITIONAL code - the two file-static SharedPtrs
(DirectGLES.cpp:82-83), the monolith arm (:361-371), NeedsRawDepthFetchSampler (:393-399)
and the pre-handle sampler-pass branch (:8015-8024). None sits behind
#if !MOBILEGL_BUILD_DISAGGREGATED; the pull build configures DISAGGREGATED=OFF and
PIPE_PUSH=OFF and links all of it, and the monolith arm is the ENTIRE body
GetRawDepthFetchSampler has there. Deleting any of it moves pull .text and fails G1's
0/0/0/0 - the rule ARCHITECTURE.md:340 states for this exact class of retired-but-compiled
code. Recorded as P13, beside the existing S7 entry in notes/p5f/f0-statics.md, rather
than attempted and reverted.

G1 on this head: .text 10822147 = 0xa52203, nm --defined-only identical to
~/w7/p7-before/pull-syms.txt (30570 symbols).
2026-09-22 17:31:51 -04:00
swung0x48 9a69e08365 [MG_IntegrationTest, docs] (P7): twelve texture and program scenarios join the split arms, with the census that put them there
P3b/P4b deepening, landing under the P7 program (PLAN-PH-P34B-P7.md 1.2, "门:纹理 /
program 场景 split 覆盖", wave 2-D package D2). Each candidate was run once under
MOBILEGL_TRANSPORT=inproc on build-split and its first blocker recorded; the row per
scenario is in notes/p34b/split-scenario-census.md.

Registered (all green on inproc, spawn and tcp): TextureView,
TextureParamsWithoutASamplerView, SampledSetStaleness, ImageSizeAfterRespec,
ImageTargetKind, IntegerBorderColor, SwizzleAccessRoutine, PostLinkAttach,
RelinkStageSet, UniformInitializer, Glsl420Declaration, IoBlockNameCollision. 71 cases
per arm, 213 new gating entries; the recorded split-arm registration count the tier-2
Magma replay walks goes 33 -> 46.

TextureParamsWithoutASamplerView settles the debt table's 待核 row: all four G9 cases
pass under all three transports, so the P6 inspection-forwarder debt is closed for this
half - and it is now re-taken on every CI run rather than by inspection.

ProgramPipelineScenario beyond *StorageBlock* is NOT registered, and that is the rule
working rather than a blocker to retire: the other nine cases are pure name/state cases
that draw nothing, and an armed DirectGLES.Split. case whose workload produced no record
is red by design (ScenarioFixture.h:84-97 - the client encoder's record ordinal must
move). Registering them would assert "records crossed" about cases with no records to
send.

DirectVulkan arms default to none, deliberately, and the note says why: the
informational tier-2 block already replays every recorded registration on DirectVulkan
under integration-magma-all-{split,spawn,tcp}, so these scenarios do get a Magma x
three-transport reading - just not a gating one. Gating them would assert P7 wave 2's
exit gate 1 from inside a P3b/P4b package, on a backend whose own @P7 refusals are still
being retired in waves A/B/C.

D1 owns PixelStoreSweep, DepthStencilReadbackMatrix, PackedWordReadback and
LayeredTextureReadback; they are not in this census.
2026-09-22 17:31:51 -04:00
swung0x48 6567bc1e16 [scripts, ci] (P7): a purity gate for Espryt's re-keyed memo families
P3b/P4b deepening, landing under the P7 program (PLAN-PH-P34B-P7.md 1.2, wave 2-D
package D2): "一条纯度 grep 门:split 构建下这些 memo 键里不得出现前端指针类型;接 CI".

scripts/ci/espryt_memo_purity.py, a sibling of link_seam_purity.py in the same style:
a hard-coded source list, a code_only() comment strip, (pattern, message) rules, and
a --self-test that runs the negative controls and then the real scan in one
invocation - which is what lets test.yml pass only --self-test and still get the
gate. Wired in next to link_seam_purity.py.

It scopes the rules to the BRACES of the families the plan names -
ResolvedTextureBindingMemo, SamplerPassMemo, UnitSamplerLookupMemo, TwinLookupMemo,
StateBackendObjectRegistry (the program registry and its five siblings) - plus the
image sweep, which is seven file statics rather than a struct and is matched by name.
Scoping matters: these two files pass frontend pointers around legitimately in
hundreds of places, and a file-wide grep would be a list of exceptions rather than a
gate.

Guard-awareness is a small conditional walker, not a preprocessor: it recognises
#if MOBILEGL_PIPE_LEGACY_MEMOS, #if !MOBILEGL_BUILD_DISAGGREGATED and the #else of
#if MOBILEGL_BUILD_DISAGGREGATED, which is every spelling these files use, and says
so in the docstring.

Six allow-list entries, each carrying its one-line reason. The two that matter:
StateBackendObjectRegistry's UnorderedMap<StateObject*, Entry> and
ResolvedTextureBindingMemo's type-erased `const void* program` are the PULL build's
only arms - deleting them moves pull .text and fails G1's 0/0/0/0, so they retire
with the pull path (P13), not here. The other four are memo VALUES rather than keys.

Seven negative controls turn red; four positive controls stay green, including a
LEGACY_MEMOS arm, the #else of a DISAGGREGATED arm, the allow-listed type-erased key
and a frontend pointer outside every memo family. Verified once more by injecting a
fresh unguarded ProgramObject* field into the real ResolvedTextureBindingMemo: the
gate reddens and names it.
2026-09-22 17:31:51 -04:00
swung0x48 5617c5c3b3 [MG_IntegrationTest, docs] (P7): the texture upload shape becomes a gate, with the view remap inside the workload
P3b/P4b R-11 / D-D4, landing under the P7 program (PLAN-PH-P34B-P7.md 1.2, wave 2-D
package D2). P4a recorded the emission shape rather than gating it because the
per-storage-owner cursor with view/owner remapping was the next phase's and gating an
unfinished shape pins a non-answer. That work landed, so the RecordProperty block is
now EXPECT_EQ against a gold row MEASURED on this cursor, and the workload carries a
third texture written through a glTextureView so the remap is inside the counted
window.

The gold row, identical on monolith / inproc / spawn / tcp:
  emit=9 box=9 rect=0 jobs=9, client emitter 9
keyed on the workload constants by static_assert. P4a's emit=6 box=6 rect=0 jobs=6
was a two-texture workload with no client emitter and is not comparable.

The client half is read from MGPipeTextureEmitter::SubDataCount() rather than from
ctu=, and that is a correction: under spawn and tcp the two roles are two processes
whose PipeStats windows close on different Present cadences (measured ctu=13 against
emit=9, and usually not yet written when the case reads), so ctu == emit is not a
well-posed comparison there. The log field is kept as a record and asserted only
where the transport says the two are window-aligned - asking the NUMBER was the first
mistake, because the server publishes a readable ctu=0 under spawn and the comparison
silently became 0 == 9. PipeStatsWindow gains LastFromClientLog/LastFromServerLog.

Split/spawn/tcp arms added as explicit per-arm blocks rather than through
mgl_itest_register_split_arms: the macro hands every arm the shared environment, and
this case needs a lane marker, PipeStats period 1, a push mask and above all a
MOBILEGL_LOG_FILE_PATH private to the arm - the library opens the log "w", so two
entries sharing a path truncate each other and the failure looks exactly like "the
counter was never emitted".

Red-once (R-16): the prescribed flip of summedArea*4 >= unionArea*3 leaves the gate
GREEN, and that is a finding. Two gates downstream dominate it - the server needs
dirtyRectCount >= 2, and UnpackRingAvailable() zeroes the rect list by design ("one
box, one job", the existing Mali-cliff mitigation). So rect=0 is the policy's
intended output, not an accident of the scatter pattern. The red-once that bites
splits the union box in two AND disables the unpack ring: box=6 rect=3 jobs=12
against the gold, while 26 pixel cases stay green. Four attempts and their reasons
are in the note.

The Mali frame-time delta D-D4 asks for beside the gold is NOT schedulable - one
Redmi, an Adreno, no Mali part - and is recorded as a declared deviation in the
scenario header and MEASUREMENTS.md 13.3 rather than modelled.

Also corrects the three places that said the client emitter had not landed: the
scenario header, the CMake probe's comment, and gates-v1.md's re-run table.
TextureEmit.h's EmitOneLevel has been incrementing it.
2026-09-22 17:31:51 -04:00
swung0x48 da30c1301e [MG_IntegrationTest, docs] (P7): the owner-keyed emission cursor gets the scenario that can see it
P3b/P4b deepening, landing under the P7 program (PLAN-PH-P34B-P7.md 1.2, wave 2-D
package D2). The code half of "按存储属主键控的发射游标与 view 索引重映射" landed in
P4a/P5e; this is the verification half it never had.

TextureViewAliasScenario, four cases: an upload through a layer-restricted view
sampled through the owner and the reverse, each once and again across draw
boundaries so the drain runs three times. Each case asserts on TWO channels at
once - sampled texels AND MGPipeTextureEmitter's SubDataCount / RefusedSubDataCount
/ DrainListSize, read through the new Harness/TextureEmitPeek. The counters are the
load-bearing half: a level that never made it onto the drain list is still uploaded
by the next whole-level respecify, by a glGenerateMipmap or by the monolith sync
path, so the picture comes out right one frame late and a pixel-only case goes green
over the regression on most trees.

Registered in the three GLES split arms, and MOBILEGL_ESPRYT_ENABLE_TEXTURE_VIEW=1
added to MGL_ITEST_GLES_{SPLIT,SPAWN,TCP,TCPDEVICE}_ENVIRONMENT - views are Espryt's
opt-in and without it every case green-skips. The Tcp arm needed the knob on the
TcpServer.Start fixture as well: that lane's server is pre-started rather than
spawned from the client, so it withheld GL_ARB_texture_view and all four cases
SKIPPED while the identical Split and Spawn arms ran.

Red-once (R-16): TextureObjectView::MarkStorageDirtyRegion made to call
PipeNoteLevelDirty on ITSELF instead of forwarding to the storage owner - the
two-cursor shape. TheViewUploadDrainRunsAgainAfterADrawBoundary goes red on both
channels simultaneously (stale owner texels AND SubDataDelta() == 0). The plan
predicted case (a); it is (c) that reds, because the first drain of a texture's life
carries the whole level anyway - recorded in the note.

TheOwnerUploadDrainRunsAgainAfterADrawBoundary is EXCLUDED BY NAME from the split
arms: it is red on all three transports and green under monolith, because an
owner-side resource_subdata moves only the owner's record.Serial and nothing
propagates that to the records viewing it. Cause and the two candidate fix sites are
named in the CMake comment and in notes/p34b/espryt-d2.md. The name stays in every
lane (G14).
2026-09-22 17:31:51 -04:00
swung0x48 ae3f43ab56 [docs] (P7): ID-P7-24 - package D1 landed, and the readback half was already closed 2026-09-22 16:42:34 -04:00
swung0x48 ba81453f19 [MG_Backend, MG_IntegrationTest, docs] (P7): the readback family runs on the split arms, and it was already green
P3b/P4b espryt D1 slice 4, landing under the P7 program (notes/p7/PLAN-PH-P34B-P7.md
§1.2 row "回读 / pack state", §3 wave 2-D).

(a) THE CENSUS, AND IT CAME BACK CLEAN. Twenty-two cases per arm now run on the three split
arms: the two F1 pack-state cases, PixelStoreSweepScenario, FramebufferChurnScenario,
PackedWordReadbackScenario, LayeredTextureReadbackScenario and
DepthStencilReadbackMatrixScenario. The row expected reds and said so ("Expect reds - that
is the point"); the first run was 65/65 before any of this commit's code changes. That is
the finding: the 回读 half was assumed open because nothing had ever run it under a
transport, and the assumption was what was wrong - ScatterTightReadbackIntoPackState and the
server's neutral-pack read were already carrying it. No case needed a fix and no case needed
a named deferral.

DepthStencilReadbackMatrixScenario gets a per-arm block of its own, because the registration
macro cannot carry MOBILEGL_ESPRYT_FORCE_DS_READBACK_EMULATION=1 and without that pin the
scenario is unfalsifiable here: ES has no core depth/stencil readback, Mesa accepts the
reads anyway, and on llvmpipe every case goes green through a path the device does not have.
Its SeparateDepthAndStencilAttachmentsAreBothReadable SKIPS on all three arms and on monolith
too, naming the driver reason itself - a driver fact, not a transport one.

(b) THE TWO std::abort EMULATIONS, checked rather than assumed. copy-image-shadow-mirror
ALREADY cannot abort: MirrorCopyImageIntoDestinationShadow returns under a transport before
reaching MGPipeUnmigratedEmulation, and has since P5b package i1. get-tex-image-shadow is
retired here, and the reason is that its own premise did not hold: P4a wrote "a split server
holds no shadow to convert", but the level storage this converts IS the server's own, fed by
the staged texture store. When MapMipmapData answers, the conversion reads bytes that exist
on this side. So it converts when the shadow is there and DECLINES BY NAME when it is not
(rule I: MGLOG_E_ONCE + GL_INVALID_OPERATION, destination untouched). std::abort inside a
server answering a client's readback takes the session down for a level it happens not to
hold, and bypasses Session::Fail, which is the funnel the census gate watches. The funnel
keeps its teeth for the other four sites - generate-mipmap-storage,
generate-mipmap-cpu-fallback, generate-mipmap-cpu-filter, texture-remint-pull - which really
do reach into a client address space; fatal_census is unchanged at 79 abort sites because
the abort lives in MGPipeUnmigratedEmulation and that function stays. Stated plainly in the
notes: no case in the armed census reaches this site under a transport (the server answers
GetTextureImage with ReadTextureImageWire, never through the backend's GetTexImage slot), so
the change is defence in depth and has no integration red-once of its own.

(c) THE CASE NOTHING COVERED, and this slice's red-once:
PixelStoreSweepScenario.ReadPixelsIntoAPackBufferHonoursPackStateAndLeavesTheGapsAlone -
glReadPixels into a bound GL_PIXEL_PACK_BUFFER at a non-zero buffer offset with
PACK_ALIGNMENT/ROW_LENGTH/SKIP_ROWS/SKIP_PIXELS all non-default, read back with
glMapBufferRange. It is the only thing that drives EmitTables.cpp's `if (pbo)` branch, the
client servicing the pack PBO itself after the reply lands; the existing sweep case reads
into a client pointer, which is the other half and has unit cover. Every number makes a term
matter: ROW_LENGTH 5 at 4 bpp is 20, which ALIGNMENT 8 rounds up to a 24-byte stride;
SKIP_ROWS 1 and SKIP_PIXELS 2 put the first pixel at image byte 32; the read is issued at
buffer offset 16, so it lands at 48. The whole buffer starts 0xCD and the case asserts both
halves - the rows land where the pack state says, and every other byte is still poison.

Red-once (R-16): dropping just the buffer-offset term from EmitTables.cpp's `start` reds the
three split entries with "readback[at] Which is: '\xCD' (205)" - the destination still poison
because the rows landed 16 bytes early - while the ambient monolith entries pass, since the
driver services its own PBO there. EmitTables.cpp is not part of this commit.

(d) SplitReadbackPackBuffer's comment claimed "The client refuses PACK_BUFFER before
emission". It does not: it KEEPS the PBO and services it after the reply, and a refusal would
have been observable as a GL error. The conclusion (the server must see no pack buffer) was
right and the reason was backwards; rewritten to ID-49's actual split of the work.

Gates: integration-split 225/225 (base 186), integration-spawn 141/141 (base 102),
integration-tcp 144/144, ctest -L unit 2408/2408, integration-magma-split 73/73,
spawn_lane_parity / fatal_census (79 sites, 0 unmarked) / link_ratchet (186, unchanged)
green, G1 .text 0xa52203 with both nm lists byte-identical to ~/w7/p7-before.
2026-09-22 16:41:54 -04:00
swung0x48 834a8917ff [MG_Backend, docs] (P7): a transport with no record arm stops by name instead of dying mid-frame
P3b/P4b espryt D1 slice 5, landing under the P7 program (notes/p7/PLAN-PH-P34B-P7.md
§1.2 row "g_fboTextureSyncList", §3 wave 2-D).

ResolveFramebufferSubsystemArm and its three siblings classify a family's arm as Handles,
Legacy or NoArm and stop only on NoArm. Under an active transport Legacy is not a slower
answer, it is a postponed crash: all four pre-handle arms reach frontend state through
accessors that are BARRIER_PULLED rows (FieldOwnership.def), and under a transport that
state belongs to a client that is no longer parked behind this apply, so the accessor's
honest answer is Fatal{UnmigratedPipeInput, "<field>@<verb>"} on the first verb that reaches
it. That is the audit's own claim about g_fboTextureSyncList - "separation-safe, because the
record arm returns first" - which holds only while the record arm is ARMED, and nothing
checked that.

Each of the four now stops, under #if MOBILEGL_BUILD_DISAGGREGATED, when the transport is
not monolith and the verdict is not Handles, through one shared
StopOnTransportWithoutRecordArm(bit, detail) that formats both causes and hands them to
BufferImpl::StopOnArmlessPipeSubsystem - this file's existing single voice for "the
configuration left no arm to run". A transport with the record arm off IS that condition
with a different cause: the legacy arm is present but unreachable. Reusing the voice keeps
the census's family word and puts the cause in the message rather than in a new word the
census would have to learn.

Said precisely, this is a RESOLVE-TIME stop and not a startup one. The four resolvers are
function-local statics resolved lazily at first use, deliberately - Managers.cpp's own
comment says a stop inside eglMakeCurrent reads to the harness as "no usable GPU" and skips
- so the stop still lands in a scenario body where ctest reports it. What it no longer does
is wait for a verb to dereference the client's state and then blame the accessor.

Red-once (R-16), and it is the safety argument the audit asserted but never ran:
MOBILEGL_PIPE_PUSH=0x1dff (the Handles lane's 0x1fff minus 0x200,
kMGPipeSubsystemFramebuffer) with MOBILEGL_TRANSPORT=inproc, running
ClearThenReadPixelsScenario.ClearWithNoDrawIsVisibleToDefaultFramebufferReadPixels. Both
runs abort; what changes is what the abort says. Without the stop:
Fatal{UnmigratedPipeInput, "GetFramebufferBindingSlot@Clear"} - mid-frame, naming an
accessor and a verb, with nothing pointing at the mask. With it:
Fatal{PipeLegacyMemosDisabled, "MOBILEGL_TRANSPORT is not monolith and
kMGPipeSubsystemFramebuffer (bit 9) is clear (or refused), so the record arm is off and the
legacy arm is unreachable: ..."}.

Gates: integration-split 202/202, integration-spawn 118/118, integration-tcp 121/121,
ctest -L unit 2408/2408, integration-magma-split 73/73, spawn_lane_parity / fatal_census (79
sites, 0 unmarked) / link_ratchet (186, unchanged) green, G1 .text 0xa52203 with both nm
lists byte-identical - the four blocks are inside MOBILEGL_BUILD_DISAGGREGATED and the pull
build compiles none of them.
2026-09-22 16:41:54 -04:00
swung0x48 767e6c8084 [MG_Pipe, MG_Test, docs] (P7): the reverse channel is nine callbacks, and the prose says what the tree does
P3b/P4b espryt D1 slice 3, landing under the P7 program (notes/p7/PLAN-PH-P34B-P7.md
§1.2 row "XFB scatter" item 3; CONTRACT-P7.md §139 already records the deletion).

Plan B section 7.1 named ten reverse-channel callbacks. The tenth, OnXfbScatterReady,
belonged to a design the tree does not have: plan B put the XFB scatter on the CLIENT - the
server would hand back the packed scratch and the client would run the patch loop over its
own shadow - so something had to tell the client the layout. P5c/P5f went the other way,
and the declaration was left behind with zero producers, zero consumers and no EventKind of
its own, while the static_assert beside kMGPipeCallbackCount made it look load-bearing.
Deleted; kMGPipeCallbackCount 10 -> 9.

Three pieces of prose that still described the old design go with it. MGPipeCallbacks.h's
own header said "ten callbacks", and now says nine and why the tenth went.
MGPipeTypes.h's MGPXfbAccounting comment said the scatter "is a read-modify-write of the
client's shadow and lives there"; it lives on the server's staged shadow, and nothing about
the scatter crosses in that record. ARCHITECTURE.md §8.5, titled "XFB scatter 搬到 client",
is rewritten to what the tree does: the pre-capture bytes are the server's own staged
shadow (rule C, since MappedData() on the apply thread is Fatal{RoleViolation,
"buffer-legacy-arm"}), the reconciled whole range comes back as one OnBufferWriteback
sliced against SEG_EVENT's capacity (slice 2), and an orphaned target is a legal one (slice
1). Its row in §8.1's callback table goes too - leaving it would be a new staleness rather
than an inherited one.

PipeCatalogue.ReverseChannelHasTenCallbacks now asserts 9. The CASE NAME IS DELIBERATELY
NOT RENAMED: G2/G14 say the ctest name set only ever grows, so renaming would delete a name
the gate watches. The count it asserts is what has to be right and the comment carries the
number.

Red-once (R-16): with the member and the count restored the case fails with
"kMGPipeCallbackCount Which is: 10 / 9u Which is: 9", which is what makes the struct's
shrink a measured fact - the static_assert beside the constant only proves the two agree
with each other.

G1 matters for this slice in particular, because MGPipeCallbacks.h is included from
MG_Backend and the struct is NOT behind MOBILEGL_BUILD_DISAGGREGATED, so
sizeof(gMGPipeCallbacks) really does shrink by 8 bytes in the pull build. The shrink is in
.bss and nothing indexes the table by count: .text is still 0xa52203 and both nm lists are
byte-identical to ~/w7/p7-before.

Gates: integration-split 202/202, integration-spawn 118/118, integration-tcp 121/121,
ctest -L unit 2408/2408, spawn_lane_parity / fatal_census / link_ratchet (186, unchanged)
green.
2026-09-22 16:41:54 -04:00
swung0x48 96fa569250 [MG_Backend, MG_Pipe, MG_Remote, MG_IntegrationTest, docs] (P7): the XFB writeback event is sliced, so a large capture stops killing the server
P3b/P4b espryt D1 slice 2, landing under the P7 program (notes/p7/PLAN-PH-P34B-P7.md
§1.2 row "XFB scatter", §3 wave 2-D).

Both XFB writeback producers posted ONE unbounded OnBufferWriteback covering the whole
bound range - ScatterCapturedRecords from its staged vector, ReadbackCapturedRanges from a
live glMapBufferRange mapping. ServerOnBufferWriteback turns each into one SEG_EVENT record
of sizeof(EventBufferWritebackHead) + bytes, and RingProducer::Reserve refuses any record
above Capacity()/2 outright - not "when full" but always, because above half a capacity a
record is placeable at some head offsets and not at others and waiting for room would be a
hang. EventRingBytes defaults to 256 KiB, so any capture target wider than ~128 KiB was
Fatal{EventRingOverflow} and the SERVER PROCESS DIED, on a buffer size, with nothing else
wrong. The producers' own Fatal text already said what to do: raise MOBILEGL_IPC_EVENT_KB
or slice the event at its producer, "as the writeback path already does" - which was true
of the client's request path and of nothing on this side.

MG_Pipe now carries the one number, published the way gMGPipeSegmentResolver is published:
gMGPipeEventRingCapacityBytes, installed by ServerSession::Accept beside the four reverse
producers and released at Close only if it is still ours, and
MGPipeBufferWritebackSliceBytes() which turns it into a per-record width - a quarter of the
ring with a 4 KiB floor, the same number and the same reasoning as the client-linkage
BufferWritebackSliceBytes one ring over. The backend can reach neither MG_Remote::Client
(wrong linkage) nor ServerSession (wrong layer), so the hook is the seam; 0 means "do not
slice", which is monolith, where the callback is a direct call with no ring under it.

Both producers loop against ONE helper, XfbWritebackSliceStep, because a second spelling of
the width is how one of them ends up still killing the server. The readback site slices
inside its mapping and still unmaps once; the scatter site slices the staged vector and
keeps the single glBufferSubData whole, since the ES store has no ring and cutting it would
change how many backend calls one application call makes. In-order delivery makes the last
slice's landing imply every earlier one, so the client needs nothing new.

Red-once (R-16): XfbAfterClipDistanceScenario.{AScatteredCapture,APlainCapture}
IntoAMegabyteRangeReachesTheApplication - a 1 MiB bound range, about eight times what one
record can ever be, with the same six capture records; glBindBufferBase over the whole
store is what makes the mirrored range a megabyte. Two cases for the two producers. With
the slicing reverted all six split entries go red on Fatal{EventRingOverflow}
("kEventBufferWriteback needs 1048600 bytes and SEG_EVENT caps ONE record at 131072") -
aborting the test process under inproc, losing the peer under spawn and tcp - while both
monolith controls stay green.

XfbAfterClipDistanceScenario and XfbRepeatedCaptureScenario now join the three split arms
in full; slice 1 could register only OrphanedCaptureTarget* because the rest could not
survive the unsliced event. XfbRepeatedCaptureScenario's gl_NextBuffer compacted-target
shape is the first thing outside the census to exercise the archive-served stride and
varyings (ProgramArchive->Link's xfb fields) on the server, and it passed on all three arms
with no further change.

Gates: integration-split 202/202, integration-spawn 118/118, integration-tcp 121/121,
ctest -L unit 2408/2408, the monolith and split XFB families 120/120, spawn_lane_parity /
fatal_census (79 sites, 0 unmarked) / link_ratchet (186, unchanged) green, G1 .text
0xa52203 with both nm lists byte-identical to ~/w7/p7-before.
2026-09-22 16:41:54 -04:00
swung0x48 faf0a1f902 [MG_Backend, MG_IntegrationTest, docs] (P7): an orphaned XFB capture target keeps its varyings
P3b/P4b espryt D1 slice 1, landing under the P7 program (notes/p7/PLAN-PH-P34B-P7.md
§1.2 row "XFB scatter", §3 wave 2-D).

ScatterCapturedRecords is a read-modify-write of the capture destination's pre-capture
bytes - a gl_SkipComponents hole has to come back holding whatever the application left
there - and under a transport those bytes are the server's staged shadow rather than a
frontend MappedData(). The split arm read a null shadow as "this target has no
pre-capture bytes, discard it". A null shadow is not an error: MGPResourceDesc's
HasDefinedContent == 0 is the store the application ORPHANED with glBufferData(size,
NULL), and Managers.cpp's own M-3 comment says every byte it has not staged since is
undefined by its own declaration. So the continue threw a whole capture away for the
ordinary streaming idiom (orphan, upload part of it, capture into it), leaving the
application's buffer holding its pre-draw bytes under GL_NO_ERROR. The partly uploaded
shape did not even reach the continue: it has a shadow, so it asked RequireStagedCoverage
for the whole bound range and died on Fatal{StageSnapshotTooNarrow,
"xfb_scatter_pre_capture"}.

Monolith has always been right here and is right by construction: `Vector<Uint8>
staged(rangeBytes)` is value-initialised, so scattering over an unreadable source scatters
over zeroes and uploads them, which is byte for byte what a fresh MappedData() hands back.
The split arm now skips only the Memcpy, never the target. Discard-with-log is kept for
splitResource == nullptr alone, which really is impossible and has no backendId to upload
into either. RequireStagedCoverage is guarded by ResourceContentIsDeclared, exactly as
Managers.cpp's pool-reuse ladder guards its own whole-store MGL_SERVER_STAGED_REQUIRE: a
declared store still Fatals on a coverage gap, an orphaned one does not.

ResourceContentIsDeclared moves out of Managers.cpp's anonymous namespace to namespace
scope beside RequireStagedCoverage and is declared in Managers.h. The two callers must not
be able to answer "did the application declare these bytes" differently. Both live inside
#if MOBILEGL_BUILD_DISAGGREGATED, so the pull build compiles neither.

Red-once (R-16): XfbAfterClipDistanceScenario.OrphanedCaptureTarget{WithAPartialUpload,
WithNoUploadAtAll}KeepsItsVaryings, registered on the three split arms under the Xfb.
prefix. With the fix reverted the six split entries go red - the partial-upload three on
Fatal{StageSnapshotTooNarrow, "xfb_scatter_pre_capture"} ([0, 696) of a buffer whose
staged coverage does not include it), the whole-orphan three on a captured varying that is
still zero - while the two ambient monolith entries pass, which is what makes it a
transport fact. Details and the lane numbers in docs/Disaggregated/notes/p34b/espryt-d1.md.

The narrow filter is deliberate: the rest of XfbAfterClipDistanceScenario is a DirectVulkan
allocation-pattern regression that has never run under a transport, and it joins the arms
in slice 2 once the writeback event it posts is sliced.

Gates: integration-split 188/188 (186 + 2), integration-spawn 104/104, integration-tcp
107/107, ctest -L unit 2408/2408, the monolith DirectGLES XFB + readback families 66/66,
spawn_lane_parity / fatal_census / link_ratchet (186, unchanged) all green, G1 .text
0xa52203 with both nm lists byte-identical to ~/w7/p7-before.
2026-09-22 16:41:54 -04:00
swung0x48 09ba88e70d [MG_Remote, docs] (P7): CONTRACT-P7 §3.2 records package B (readback race, copy-image, default blit, depth mip bake, mip diagnostic), ID-P7-23 2026-09-22 16:40:02 -04:00
swung0x48 76b7469389 [MG_Backend, MG_IntegrationTest, docs] (P7): the wire readback waits for the frame it is reading, instead of inferring who wrote it
Exit gate 3, OpenRA on the Redmi. The mechanism, in two lines of this tree:

  WireFramebuffer.inc, ReadWirePixels: `if (HasPendingRecordedWork() &&
  !FlushPendingCommands())` - FlushPendingCommands (VulkanRenderer.cpp:13534-13598)
  puts the frame's draws on the queue under a POOLED FENCE OF ITS OWN and returns
  without waiting; there is no vkWaitForFences on its success path. The
  vkCmdCopyImageToBuffer that follows is recorded into a fresh command buffer and
  becomes a LATER submission.

  VulkanRenderer.cpp:10707-10748, SubmitReadbackCommandsAndWait: it waited ONE fence,
  its own, and then called OnSubmitsCompletedUpTo(frame.lastSubmitIndex) - which
  RECORDED the earlier submission as complete and retired its fence, so afterwards
  nothing could observe that it had not been waited for. The comment beneath it said
  "the wait proved every submission complete". That is true of the monolith arm, which
  never flushes before a readback and therefore has its draws and its copy in one
  submission under one fence. It was not true of the wire arm.

The only thing tying the two submissions together was the pipeline barrier that the
readback's TransitionWireImage records - and that barrier derives its SOURCE scope from
the image's tracked layout, through GetImageTransitionSourceState
(VulkanRenderer.cpp:1784-1825). Two of its arms, VK_IMAGE_LAYOUT_UNDEFINED and
VK_IMAGE_LAYOUT_PRESENT_SRC_KHR (:1787-1791), return VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT
with no access bits, which is Vulkan's spelling of an EMPTY first scope: wait for
nothing. A just-presented default framebuffer is in the second of those layouts. The
slower pipeline creation is, the longer the draws' submission runs, and the more of the
frame's LAST-drawn objects the copy can miss - which is what the device's diff image
shows.

Measured on this host (OpenRA, DirectVulkan, inproc), at the moment the copy is
recorded: submit=72 completed=71. The submission carrying the whole frame's draws is in
flight and unwaited. That is structural, not a coin flip, and it reproduces every run.

The fix is two halves, both wire-only:

 - RecordWireReadbackWriteBarrier (WireFramebuffer.inc, new): the readback records its
   own ALL_COMMANDS -> TRANSFER, MEMORY_WRITE -> TRANSFER_READ|MEMORY_READ barrier
   before the copy rather than inferring from a layout who it should wait for. A
   barrier's first scope is every command earlier in SUBMISSION order on the queue, so
   it reaches across the submission boundary.
 - SubmitReadbackCommandsAndWait now waits every in-flight fence at or below its own
   submission index. Under #if MOBILEGL_BUILD_DISAGGREGATED: the pull build has no wire
   arm, cannot flush before a readback, and is the image G1 pins - its single-fence wait
   is unchanged. The cost is nothing a readback does not already pay.

THE COORDINATOR'S REPRODUCTION PATH DOES NOT WORK ON THIS HOST, and that is recorded
rather than glossed. MOBILEGL_TEST_PIPELINE_CREATE_DELAY_MS is added as asked
(split-only, inert unless set, on the pipeline-creation MISS path in
GetOrCreatePipelineWithInput) and it does bite - the OpenRA DirectVulkan SPLIT retrace
goes from 1.88 s to 4.53 s at 25 ms - but ssim stays 1.000000. The reason is in the
probe above: this host's default framebuffer is tracked as VK_IMAGE_LAYOUT_GENERAL at
the readback, which takes GetImageTransitionSourceState's `default:` arm (ALL_COMMANDS /
MEMORY_WRITE) and orders the copy by accident. The device's is not GENERAL.

Red once (executed, reverted) is therefore a COUNT, not a picture, and it is
deterministic: instrumenting how many fences the readback actually waits on gives
fences=2 upTo=73 completed=71 with the fix (submission 72, the draws, plus its own 73)
and fences=1 without it, with completed=71 either way - i.e. 72 provably outstanding and
provably unwaited before.

Lane: three readback-after-work cases (VertexIdSamplerAndScalarUniformPixels,
ComputeImageStoreFramebufferPixels, GenerateMipmapPixels) registered on the three Magma
arms under MOBILEGL_TEST_PIPELINE_CREATE_DELAY_MS=25, prefix .PipelineDelay. They are a
REGRESSION GATE, not the reproduction: they pin that a readback still returns the draw's
pixels when every pipeline creation on the path costs 25 ms.

Gates: integration-magma-split 84 (+3), -spawn 63 (+3), -tcp 65 (+3), 0 red;
ctest -L unit 2410; informational integration-magma-full-split 516, 0 red, 59 skip;
fatal_census green; link_ratchet --assert-monotone unchanged at 186, p7-magma 101;
G1 pull .text 0xa52203, defined-symbol name list identical (30570); OpenRA DirectVulkan
SPLIT retrace ssim 1.000000 with and without the delay, two runs each.

Owed to the device: the three cold-cache repeats. If they still diverge, the next two
suspects are named in notes/p7/magma-b.md §5.6 - the eleven silent `return false` drop
sites in SetupWireDraw (one of which is `pipeline == VK_NULL_HANDLE`, on the same path
as the device log's three "handing vkCreateShaderModule an INVALID module" errors), and
SetImageContentDefined being set BEFORE the draw is recorded on the wire arm while the
monolith sets it from EndRenderPass, which silences the "no write is recorded" warning
exactly when it should fire. Neither is in this commit's scope.
2026-09-22 16:37:54 -04:00
swung0x48 583e12de76 [MG_Backend, MG_IntegrationTest, docs] (P7): the default-framebuffer blit stops refusing shapes the arm beneath it can do, and the 3D bug that hid behind the refusal is fixed
CONTRACT-P7 §3.2's `default-color-blit-shape@P7`, retired in three pieces, because one
Fatal was covering three unrelated situations.

The shader pass it guarded is there for ROTATION and nothing else: it can only SAMPLE
its source - sampler2D, a float result, one of two 2D view types - so an RGBA8UI
attachment, a 3D slice or an array layer made it return false. The refusal then fired
EVEN ON A SURFACE WITH NO ROTATION, where the ordinary vkCmdBlitImage arm twenty lines
below would have done the blit and does handle those shapes.

 - identity: fall through to that arm;
 - a non-identity rotation with a source the pass cannot sample: move the region into an
   owned 2D single-sample image of a sampled-float format in the SAME size-compatibility
   class and shader-blit that;
 - outside the four rotations (a mirror, or a mirror-rotation): decline by name, saying
   which transform. Neither arm can express a reflection and this package has no device
   that reports one to check a result against.

Reinterpretation is the point for the integer case and GL agrees: glBlitFramebuffer
between an integer and a normalized format is INVALID_OPERATION (4.6 core 18.3.1), so
there is no defined result to be wrong about - only a shape that used to end the session
and now produces the source's bits. For a float source of the wrong view type the copy
is a plain slice extraction and the result is exactly right.

THE RETIREMENT UNCOVERED A REAL BUG, which is the part worth reading. The ordinary blit
arm addressed a 3D source as `{aspect, level, source.layer, 1}` with z 0..1. A 3D image's
slice rides Z; its only subresource is layer 0 - which is how the copy endpoint's
CopyImageSliceMapping and the multisample resolve arm above already address one. So a
blit from slice 1 of a 3D texture read slice 0, out of a layer the image does not have.
It was invisible because that shape hit the Fatal first and never reached the arm: the
new scenario's first run had the scratch arm correct and the fall-through arm wrong in
all four quadrants, which is how it fell out. Both arms agree now.

The scenario's source is slice 1 of a 3D texture - ordinary GL, and exactly the shape the
shader pass cannot sample - with slice 0 filled in a colour that appears in no quadrant,
so a scratch copy that forgot the z origin would still be reading real texels and had to
be made distinguishable. It is registered twice on each arm: plain, taking the identity
fall-through, and under MGITEST_MAGMA_FORCE_DEFAULT_BLIT_SCRATCH, taking the rotated
arm's scratch path on an identity surface, where it must produce THE SAME PIXELS. That
equality is the assertion, and it is the only way the scratch arm is reachable from a
host lane at all: every surface this lane can create reports
VK_SURFACE_TRANSFORM_IDENTITY, as does the Redmi pbuffer this cluster's evidence comes
from. The knob does not fake a rotation - nothing downstream is told the surface turned.

Red once (executed, reverted): with MagmaWireFatal("default-color-blit-shape@P7")
restored, both entries die by that name; without it, 2/2 green.

Gates: integration-magma-split 81 (+2), -spawn 60 (+2), -tcp 62 (+2), 0 red;
ctest -L unit 2410; informational integration-magma-full-split 516, 0 red, 59 skip;
fatal_census green; link_ratchet --assert-monotone unchanged at 186, p7-magma 101;
G1 pull .text 0xa52203, defined-symbol name list identical (30570).

Owed to a device window: the scratch arm's geometry under a REAL ROTATE_90/180/270. The
knob proves it produces the fall-through's pixels on an identity surface, which exercises
every line of it except the rotation constant the vertex shader applies.

WireColorBlit.inc and VulkanRenderer.h are edited outside this package's named file list.
Both belong to this cluster (the colour blit's implementation, and four private
declarations beside the existing blit ones); flagged for the integrator.
2026-09-22 16:37:54 -04:00
swung0x48 4f857aa5c6 [MG_Backend, MG_IntegrationTest, docs] (P7): a same-image copy takes one GENERAL transition instead of the session
CONTRACT-P7 §3.2's `copy-image-in-place@P7`, retired.

glCopyImageSubData between two subresources of ONE texture is ordinary, defined GL -
4.6 core 18.3.2 spells out that a texture and a view of it are two objects over one
image and may be copied between. The wire arm took the session down for it, and the
reason it gave was true of the code beneath it and of nothing else: the
TRANSFER_SRC/TRANSFER_DST pair cannot describe one image, because both endpoints share
one tracked VkImageLayout and the second transition undoes the first.

VK_IMAGE_LAYOUT_GENERAL can describe it - both sides of vkCmdCopyImage accept it. So a
same-image copy now takes ONE transition of the whole image to GENERAL (the source level
and the destination level are both inside it), GENERAL on both sides of the command, and
ONE restore, with an access mask carrying both directions because that one barrier is
the edge for both. The monolith arm still declines every in-place copy a few lines up,
so the wire arm is the one that performs this - which is what §3.2 asks for rather than
parity. `inPlace` is constexpr false in the pull build so every branch it guards folds:
this function is in the monolith image and G1 pins that image's .text.

The one shape that stays a decline is same image + same level + overlapping slices +
overlapping rectangle, all three at once. GL leaves that undefined, and GENERAL makes
the command legal to RECORD rather than meaningful to execute - vkCmdCopyImage orders
nothing between its own reads and writes inside a region, so a tiler produces whichever
half of each texel it reached first. Level 0 to level 1, disjoint layers, and disjoint
rectangles of one level are all defined, and all performed.

Two split-only cases on the three Magma arms, over one 8x8x2 two-level array texture
whose every touched subresource is a different colour - a flat fill would let "copied to
the wrong level", "copied to the wrong layer" and "did not copy" all read as correct.
CopyImageInPlaceAcrossLevelsAndLayers does level 0 -> level 1 and then layer 0 -> layer 1
and checks the untouched subresources too; CopyImageInPlaceOverlapDeclinesAndLeavesTheLevelAlone
copies (0,0)-(4,4) onto (2,2) and pins every texel unchanged - an executed copy would
turn the texel at (4,4) from yellow to red, so the assertion can tell.

Red once (executed, reverted): with MagmaWireFatal("copy-image-in-place@P7") restored,
both entries die with Fatal{UnmigratedVerb, "Magma:copy-image-in-place@P7"}; without it,
2/2 green.

Gates: integration-magma-split 79 (+2), -spawn 58 (+2), -tcp 60 (+2), 0 red;
ctest -L unit 2410; informational integration-magma-full-split 515, 0 red, 59 skip;
fatal_census green (79 sites / 43 words / 0 unmarked); link_ratchet --assert-monotone
unchanged at 186, p7-magma 101; G1 pull .text 0xa52203 and the defined-symbol name list
identical (30570) - the constexpr-false branches fold as intended.

This tree's integration TCP endpoint moved to 127.0.0.1:40913: a sibling worktree's
libMobileGLServer was already listening on pipe's default 40613 and took the whole tcp
lane down at TcpServer.Start. It is a local configure option, not a lane semantic.
2026-09-22 16:37:54 -04:00
swung0x48 fdd611e83d [MG_Backend, MG_Test, MG_IntegrationTest, scripts, docs] (P7): the depth mip chain is baked, and every baked header in the tree now has to match its source
CONTRACT-P7 §5.1's (A) and (D), and (D) is what makes (A) safe to have.

(A) WireDepthMipmap.{vert,frag,inc}. VK_FORMAT_FEATURE_BLIT_DST is OPTIONAL for depth
formats, so a depth chain through a shader is the ordinary case and not the exotic
one - the monolith arm has carried such a fallback since before the split. The wire
arm had none, so a depth texture the device would not blit died on the first
disjunction of `mipmap-shader-format-or-shape`, and a DEPTH|STENCIL one died a guard
earlier as `depth-stencil-mipmap@P7` whether or not it could be blitted.

The monolith's fallback is not reusable, which is the whole of (A): it builds a GL
ProgramObject, two ShaderObjects and a GL sampler, binds through the uniform manager's
global UBO and writes uSrcTexelSize as a default-block uniform - four frontend objects
and a GL compiler on the server's apply thread. The baked program is the same shader
with the same arithmetic on a push constant and its own descriptor set. The fragment
stage is a PORT and not a rewrite: §3.2 retires a refusal by making the wire arm
observably the monolith's, so the 2x2 box lands where the monolith's lands, including
the half texel its ivec2(texCoord * srcTexelSize) sits off the destination centre.

`depth-stencil-mipmap@P7` therefore splits. DEPTH-only retires into the baked program.
DEPTH|STENCIL declines - MGLOG_E_ONCE and return, before the chain is grown, so a level
this verb will not fill is not created and then recorded as defined. A decline rather
than a Fatal because the monolith does no better: its fallback asserts depth-only, and
an assert is nothing in the release build both arms ship as. Nothing in GL promises a
stencil mip either, and this pass writes gl_FragDepth - it would leave the stencil
aspect of the level it just produced undefined.

(D) MOBILEGL_BAKED_INTERNAL_SHADERS. A baked header is a compiler output committed with
its input beside it and nothing connecting the two: edit the source and the build stays
green while the shader that runs is the one from before the edit. MG_Test's
BakedInternalShadersTest recompiles all eight arrays - WireColorBlit x2,
WireDepthMipmap x2, the iterationRP witness, the three probe stages - from their
committed GLSL with the in-tree glslang and compares word for word.

THE GATE IS THE BAKER. scripts/bake_internal_shaders.py compiles nothing; it runs this
binary with MOBILEGL_BAKE_INTERNAL_SHADERS and splices the words it emits. A gate that
used the glslang LIBRARY while the baker shelled out to a glslangValidator BINARY would
be two compilers with no rule about which is right, and the day their defaults diverge
the gate reds on a header nobody can regenerate.

What the first run measured, which is the one thing nobody could know beforehand:
WireColorBlitSpirv.h and all three arrays of PrimitivesGeneratedNoXfbProbeSpv.h were
already fresh - the hand-run glslangValidator words reproduce exactly. The iterationRP
witness was stale, as its own prose said; it is regenerated here (2142 -> 2143 words),
and the witness magic 0x50323033 is still there because it is a constant in the .comp
rather than a property of the words. All three provenance comments now point at the
script and the gate instead of at a tool on somebody's PATH.

Scenarios, on the three Magma arms: GenerateMipmapDepthWithoutNativeBlitPixels runs
DEPTH_COMPONENT24 and DEPTH_COMPONENT32F under MGITEST_MAGMA_FORCE_SHADER_MIPMAP=1 and
reads level one back through a real sampled attachment, asserting per half and leaving
row 3 - the straddle band - unpinned, because pinning it would assert the half-texel
offset rather than the mip. GenerateMipmapDepthStencilDeclinesAndKeepsTheSession takes
no knob (the decline is on the aspect, before blit support is asked about) and pins that
the session survives and level zero still holds what was cleared into it.

Red once, three of them, executed and reverted:
 - a one-character edit to WireDepthMipmap.frag (no rebuild - the gate reads the source
   at runtime) reds the WireDepthMipmapFragment row at word 279;
 - with the depth arm disabled, the DepthMip entries die with Fatal{UnmigratedVerb,
   "Magma:mipmap-shader-format-or-shape format=125 ... aspect=0x2 ...
   missing=COLOR_ATTACHMENT"} - format 125 is X8_D24_UNORM_PACK32, which slice 1's
   diagnostic names for us;
 - with the Fatal restored, the decline case dies as
   Fatal{UnmigratedVerb, "Magma:depth-stencil-mipmap@P7"}.

Gates: integration-magma-split 77 (+2), -spawn 56 (+2), -tcp 58 (+2), 0 red;
ctest -L unit 2410 (+8, the gate's rows); informational integration-magma-full-split
513, 0 red, 59 skip (base 510/0/57); fatal_census green (79 abort sites, 43 words, 0
unmarked); link_ratchet --assert-monotone unchanged at 186, p7-magma 101.

G1 holds and it was not obvious: the witness header is compiled into the pull build
too, and its word count changed. Pull .text is still 0xa52203 and the defined-symbol
name list is identical to ~/w7/p7-before/pull-syms.txt line for line (30570) - the
change lands in .rodata and widens no instruction encoding.

VulkanRenderer.h is edited outside this package's file list: the new pass needs four
private declarations next to the colour blit's. Flagged for the integrator.
2026-09-22 16:37:54 -04:00
swung0x48 c4be38005a [MG_Backend, MG_IntegrationTest, docs] (P7): the shader mip arm names the format it refused and renders through scratch when the image cannot be attached
Two Iris cases died on the Redmi (Adreno 830v2, DirectVulkan inproc) with
Fatal{UnmigratedVerb, "Magma:mipmap-shader-format-or-shape"}, deterministically and
under both run-ahead and lockstep, while the monolith arm and the whole host lane
stayed green. That one line was the entire evidence, and five disjunctions share the
name, so the one thing the device could have told us - which format, at what shape,
missing which feature bit - was the one thing it did not.

The name stays first and unprefixed, because the census, the retrace refusal tally
and every log grep key on the leading word; the driver's own answer follows it:
format enum, the view's storage format when it reinterprets, extent, levels, aspect,
view type, usage bits, layer range, and the names of the optimalTilingFeatures bits
that are ABSENT. The next device run reads a line instead of re-deriving it.

The refusal also loses half its territory. The shader arm rendered STRAIGHT INTO the
next level, which needs the image to carry COLOR_ATTACHMENT usage - and a texture
whose format the device declines as a colour attachment never gets that bit, so every
sampled-only texture without BLIT support fell into the refusal rather than into a
fallback. When the FORMAT can be rendered and only the IMAGE cannot, an owned scratch
attachment of the destination level's size takes the pass and vkCmdCopyImage moves the
mip in: a raw texel move, no conversion, texels still from GPU contents rather than a
staged shadow, and the scratch retires with the submit that consumes it - the shape the
multisample resolve arm above it already uses.

Neither shader arm had ever executed on the gate machine. lavapipe reports
BLIT_SRC|BLIT_DST for every colour format this binary can create, so GenerateWireMipmap
takes its native arm for all of them; GenerateMipmapPackedFloatPixels next door looks
like it covers the shader arm and does not, because R11F_G11F_B10F blits natively here
too. MGITEST_MAGMA_FORCE_SHADER_MIPMAP drops the native arm instead, one tier per shape
(1 = render into the level, 2 = render into scratch and copy back), read once and
compiled only into the disaggregated build.

F1WireScenario.GenerateMipmapWithoutNativeBlitPixels paints two different halves rather
than one flat colour - a pass that loses GL's row order averages to the same number
everywhere, and a single-colour assertion would call that correct - and asserts level
one per half. It is registered on the three Magma arms twice, ONE ENVIRONMENT EACH: the
knob changes how every mipmap in the process is generated, so a lane-wide flip would
silently retarget the four green mipmap entries beside it.

Red once (executed, reverted): with the scratch arm forced off, the tier-2 entry dies
with Fatal{UnmigratedVerb, "Magma:mipmap-shader-format-or-shape format=37 extent=8x8x1
levels=4 aspect=0x1 view=1 usage=0x17 layers=0+1/1 missing=DEPTH_STENCIL_ATTACHMENT"},
by name and with the shape in it; restored, the same filter is 8/8 green.

Gates: integration-magma-split 75 (+2), -spawn 54 (+2), -tcp 56 (+2), 0 red;
ctest -L unit 2402/2402; informational integration-magma-full-split 511, 0 red, 58 skip
(the new case skips where the knob is unset); fatal_census green (79 abort sites, 43
family words, 0 unmarked - no site added, the funnel is unchanged); link_ratchet
--assert-monotone unchanged at 186, p7-magma bucket 101.
2026-09-22 16:37:54 -04:00
swung0x48 5937b4fcf2 [MG_Remote, docs] (P7): CONTRACT-P7 §3.2 records package A (byte-tail retired, ranges clamped, alignment declines), ID-P7-21/22 2026-09-22 16:23:32 -04:00
swung0x48 2207b6d9ee [docs, MG_Backend, MG_IntegrationTest] (P7): package A's eight refusals are on record, and only real sites still carry @P7
notes/p7/magma-a.md: the per-site table (retired / declined / kept-fatal with the
reason), every red-once with its captured output, the gate readings, and what is
handed on.

The three places package A's own prose still named `uniform-buffer-byte-tail@P7`
lose the marker. They are comments, not refusal sites, and leaving the marker in
would have made a tree-wide `grep @P7` count them against packages B and C. The
census now reads 15 -> 7, and all seven are real: WireFramebuffer.inc x4 and
VulkanRenderer.cpp x1 (package B), WireDraw.inc x2 (package C).

Three findings worth the integrator's attention, all in the note:

1. Most of this cluster was unreachable, and that is provable rather than
   assumed. The frontend validates buffer-range offsets against the very caps the
   backend then re-checks - GL_{UNIFORM,SHADER_STORAGE,TEXTURE}_BUFFER_OFFSET_ALIGNMENT
   and GL_MAX_UNIFORM_BLOCK_SIZE are all published from the same Vulkan limits -
   so five of the eight sites could only ever have fired on a malformed peer. The
   two that CAN fire are `uniform-buffer-byte-tail` (GL constrains a range's
   offset but not its size) and `unaligned-atomic-buffer-range` (GL gives
   GL_ATOMIC_COUNTER_BUFFER no queryable alignment, so the frontend only enforces
   offset % 4 while the backend needs minStorageBufferOffsetAlignment).

2. fatal_census does not move for any of this, and no re-baseline is due. Wave 0
   already routed these funnels through MGPipeSessionFailHook, so the abort lives
   in FatalFunnel.cpp and retiring a WireDescriptorFatal call site changes no
   census number: 79 abort sites / 43 family words, unchanged, RC=0. The brief
   expected the count to fall because the sites were bare aborts before wave 0.

3. DirectVulkan.Split.Full.IterationRPProgram203Scenario.FixedCompleteInputProducesFixedCompleteGoldenOutput
   is red on this worktree's build-split and is NOT ours: it reproduces 3/3 with
   package A's backend diff fully reverted, and the mismatch bits are identical
   to the S1 exposure texel in magma-two-process-first-run.md. The contract books
   that one against the tcp arm, and ~/w7/pipe's own build-split passes it, so
   whether it is a split-arm regression between 460a3a72 and 610fa281 or a
   configure-dependent S1 needs the integrator.

Gates on the finished branch: integration-magma-split 75/75, -spawn 54/54, -tcp
56/56, -buffers 21/21, full-split 512 entries 454 PASS / 57 SKIP / 1 red (the
pre-existing one above; base was 453/57), ctest -L unit 2408/2408,
spawn_lane_parity RC=0, link_ratchet unchanged at 186, fatal_census RC=0, and G1
pull .text = a52203 with nm --defined-only identical to p7-before/pull-syms.txt
(added=0 removed=0).
2026-09-22 16:22:40 -04:00
swung0x48 a2d4bc9db6 [MG_Backend] (P7): an unsupported texel-buffer format reads zeros instead of killing the session
`texel-buffer-native-format@P7` becomes a decline (rule I (a)), and with it the
last refusal site in UniformManager.cpp is gone. GL's buffer-texture format list
is wider than what Vulkan mandates for texel buffers - GL_RGB32{F,I,UI} and the
16-bit UNORM members are the realistic misses on a real driver - so this is a
device capability gap, not anything the application did wrong and nothing it
could have sent differently.

The placeholder is the right answer here where it was not for the unaligned range
in 7c373d60, and the difference is whether a correct answer exists at all. There
the bytes are addressable and only the descriptor cannot name them, so a
placeholder would have hidden a fixable gap behind silently dropped writes. Here
the device cannot represent this format as a texel buffer in any form, so the
honest observable is monolith's for a missing view: the fetch reads zeros and the
draw survives. AcquireUnboundTexelBufferView already falls back to the R32 member
of the shader's numeric class, whose texel-buffer support is mandatory, so the
fallback cannot itself fail for want of device features.

red-once (R-16). lavapipe supports every format the buffer-texture cases use, so
the site is not reachable on the host - the red is the same case with the device
support check forced to fail, both shapes built and run back to back:

  DirectVulkan.{Split,Spawn}.Full.BufferTextureScenario.VertexStageTexelFetchReadsTheBufferAndTracksItsUpdates
    unforced  2/2 Passed (the format IS supported here)
    before (7c373d60 + forced)  Subprocess aborted -
                                Fatal{UnmigratedVerb, "Magma:texel-buffer-native-format@P7"}
    after  (this commit + forced)  ***Failed on the texel data and NOT on a Fatal:
                                the fetch read the zero placeholder, the session lived
    restored  2/2 Passed, no force markers left

Which format actually trips this is a device question, not a host one. Adreno is
where GL_RGB32F buffer textures are expected to miss, so the decline's real
observable wants a device pass; notes/p7/magma-a.md lists it.

Lanes: integration-magma-split 75/75, -spawn 54/54, -tcp 56/56,
integration-magma-buffers 21/21. `grep -n '@P7'` on this file now returns one
line, and it is the comment that records the A.1 retirement - no refusal sites
remain.
2026-09-22 16:22:40 -04:00
swung0x48 ae57d01168 [MG_Backend, MG_IntegrationTest] (P7): an unaligned counter range declines its binding instead of killing the session
`unaligned-atomic-buffer-range@P7`, `unaligned-storage-buffer-range@P7` and
`unaligned-texel-buffer-range@P7` become named declines (rule I (a)). The two
offset-overflow guards next to them keep their named Fatal under their own words:
those are arithmetic that cannot happen, not shapes an application can bind.

The atomic-counter one is the only member of this cluster an application can
actually reach, and it is a real live gap, not a hypothetical. GL 4.6 core 6.1.1
gives GL_ATOMIC_COUNTER_BUFFER no queryable offset alignment, so
ValidateBufferRangeOffsetAndSize only enforces offset % 4 on it - while glslang
lowers every atomic_uint onto a synthesized storage block whose descriptor offset
must be a multiple of minStorageBufferOffsetAlignment (16 on lavapipe, 64 on
Adreno). `glBindBufferRange(GL_ATOMIC_COUNTER_BUFFER, 0, buf, 4, 4)` is a legal
GL bind that this backend cannot express, and it used to take the session with
it. The plain SSBO and texel halves are unreachable by comparison:
GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT and GL_TEXTURE_BUFFER_OFFSET_ALIGNMENT
are published from the same Vulkan limits, so the frontend already answers
GL_INVALID_VALUE.

DECLINED RATHER THAN COPIED, and this is the package's one real design choice.
Copying the window into an aligned transient slice reads correctly and then
silently DROPS every shader write. This site cannot tell a writable storage block
from a read-only one - there is no readonly bit in the storage-block reflection,
an atomic counter is written by definition, and MarkWireBufferGpuWritten right
below already assumes any binding here may be written. A correct copy needs a
post-dispatch copy-back from the transient slice to the unaligned source, ordered
with a barrier, and the place to record that is the draw/dispatch tail in
VulkanRenderer.cpp / WireDraw.inc - wave-2 packages B and C, not this file.
Returning stale counters with no error is worse than losing the dispatch with a
named line in the role log. notes/p7/magma-a.md carries the recipe for whoever
holds those files.

The texel half keeps the draw where it safely can: a read-only buffer texture
falls back to the same placeholder view an unbound one gets (zeros, which is what
a missing view gives in monolith), while a writable imageBuffer declines - the
placeholder is shared by every unbound binding, so absorbing real stores into it
would alias other bindings as well as lose the writes. Copying is not available
there either: the transient arena carries no {UNIFORM,STORAGE}_TEXEL_BUFFER
usage, so no VkBufferView can be made over it, and widening a shared per-frame
allocation to buy an unreachable path is the wrong trade.

red-once (R-16), a reachable case this time, all three Magma two-process arms:

    DirectVulkan.{Split,Spawn,Tcp}.Buffers.F1WireScenario.UnalignedAtomicCounterRangeDeclinesAndTheSessionLives
      before  Subprocess aborted - Fatal{UnmigratedVerb, "Magma:unaligned-atomic-buffer-range@P7"}
      after   3/3 Passed, and the role log carries
              "block 'gl_AtomicCounterBlock_0' is bound at offset 4, which is not a
               multiple of this device's minStorageBufferOffsetAlignment (16);
               atomic-counter ranges cannot be expressed as a descriptor ... DECLINED
               (unaligned-atomic-buffer-range)"

The case pins the decline's whole observable: the bind raises no GL error, the
dispatch does not run, every byte of the counter buffer is unchanged, and the
context still serves ordinary work afterwards. A future copy-back turns one
EXPECT into seed[1] + 1 and leaves the rest standing.

Lanes: integration-magma-split 75/75, -spawn 54/54, -tcp 56/56,
integration-magma-buffers 21/21.
2026-09-22 16:22:40 -04:00
swung0x48 de049f9ea0 [MG_Backend] (P7): a range past the device's native limit is clamped, not refused
`uniform-block-native-range@P7` and `storage-buffer-native-range@P7` both become
clamps with an MGLOG_E_ONCE (rule I (a)). Neither is reachable from a conformant
program: GL_MAX_UNIFORM_BLOCK_SIZE is published straight from the same
maxUniformBufferRange and GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT /
GL_MAX_SHADER_STORAGE_BLOCK_SIZE from the same storage limits
(BackendLoaders/Vulkan/Loader.cpp), so a block this big fails to link long before
a draw. Refusing was a Fatal guarding a shape nothing can produce.

The monolith arm is the argument. It binds whatever the GL range asked for and
never consults maxStorageBufferRange at all (ResolveStorageBufferDescriptor's
resident path), so the wire arm was inventing a death the other backend does not
have. If the two limits ever disagree, the honest answer is the bytes the device
CAN name: everything past the clamp is outside the descriptor and reads back as
zero under robustBufferAccess, which is inside GL's "undefined" for a block the
implementation never promised to hold. Clamping keeps the range constant across
draws, so the descriptor-set reuse hash is unaffected.

red-once (R-16). Neither site is reachable from a test, so the red is the same
two cases with the device limit forced down to 16 bytes, both shapes built and
run back to back on the Split and Spawn arms:

  uniform-block-native-range (reflected block 32, limit forced to 16)
    before  Subprocess aborted - Fatal{UnmigratedVerb, "Magma:uniform-block-native-range@P7"}
    after   BOTH ARMS PASSED. The clamp keeps every byte the descriptor can name,
            so the case's own pixel is unchanged, and the role log carries
            "reflected uniform block 0 is 32 bytes, past this device's
             maxUniformBufferRange of 16; the binding is clamped to the limit
             and the remainder reads zero"

  storage-buffer-native-range (block 'Output' 24 bytes, limit forced to 16)
    before  Subprocess aborted - Fatal{UnmigratedVerb, "Magma:storage-buffer-native-range@P7"}
    after   ***Failed on the data and NOT on a Fatal - the session lives, 8 of the
            24 bytes are genuinely past what the forced limit can name, and the
            role log carries "block 'Output' bound 24 bytes, past this device's
            maxStorageBufferRange of 16; the binding is clamped to the limit"

  restored (this commit as committed): 4/4 green, no force markers left.

Lanes: integration-magma-split 74/74, -spawn 53/53, -tcp 55/55,
integration-magma-buffers 20/20.
2026-09-22 16:22:40 -04:00
swung0x48 362b27df8c [MG_Backend] (P7): a transient ring past 4 GiB declines its uniform binding instead of killing the session
`uniform-buffer-dynamic-offset@P7` becomes a decline (rule I (a)). A Vulkan
dynamic offset is a uint32_t, so a transient uniform ring that grew past 4 GiB
inside one frame has nowhere to put the padded block - which is our ring's
problem, not a protocol fault, and nothing the peer could have sent differently.
Killing the session for it was the wrong shape twice over.

The monolith arm has the same ceiling and truncates silently on the way out
(ResolveDynamicUboDescriptor's static_cast<Uint32>), i.e. it binds a WRONG window
rather than none. Declining is the strictly better observable of the two: one
binding is skipped, the draw is lost, the session keeps running, and the next
frame's rewound ring resolves it.

red-once (R-16). The site is not reachable from a test - a 4 GiB transient arena
inside one frame is not something an integration case can build - so the red is
the same case with the ceiling condition forced true, both shapes built and run
back to back on the Split and Spawn arms:

  before (fb25ed6b + forced condition)
    DirectVulkan.Split.Buffers.F1WireScenario.ShortUniformBufferRangePadsMissingBytes
    DirectVulkan.Spawn.Buffers.F1WireScenario.ShortUniformBufferRangePadsMissingBytes
        Subprocess aborted - MGPipe: Fatal{UnmigratedVerb, "Magma:uniform-buffer-dynamic-offset@P7"}

  after (this commit + forced condition)
    both ***Failed on the pixel, NO Fatal and NO abort, and the role log carries
    ResolveWireUniformBufferPayload: the transient uniform ring reached offset 0,
    past the 4 GiB a Vulkan dynamic offset can name; uniform binding point 3 is
    declined for this draw

  restored (this commit as committed): both green, no force markers left.

Lanes: integration-magma-split 74/74, -spawn 53/53, integration-magma-buffers
20/20. -tcp not read this round: port 40613 is held by a concurrently running
lane in another worktree, which is a fixture collision and not a result.
2026-09-22 16:22:40 -04:00
swung0x48 450d30759e [MG_Backend, MG_IntegrationTest] (P7): a uniform range that ends mid-word is copied, not refused
`uniform-buffer-byte-tail@P7` is retired. It was the live boundary the Redmi
Minecraft run reported (notes/p5f/magma-inproc-fix.md #5) and, since wave 0, a
named Fatal that took the session with it.

Only the SIZE of a uniform range can land off a word. glBindBufferRange already
forces the offset onto GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT and records
GL_INVALID_VALUE otherwise (GL_Buffer.cpp ValidateBufferRangeOffsetAndSize), and
every device reports that as a multiple of four - but GL 4.6 core 6.1.1 puts no
such rule on size, so a 19-byte bind is legal and 19 % 4 is not 0. The sub-word
start is only reachable protocol-side; the new arm handles it anyway rather than
leaving a second shape to trip over later.

CopyWireBufferSubWordRangeToSlice rounds the read of the APPLICATION's store OUT
to whole words, clamped to the store's end so the widened window can never touch
a byte the application does not own, and lands it in a transient staging slice.
Only the staging -> padded-block shift is sub-word, and that happens inside our
own arena; vkCmdCopyBuffer places no alignment rule on a region's offsets or size
(unlike vkCmdUpdateBuffer / vkCmdFillBuffer), so the shift is a plain legal copy.
The block still starts at the padded slice's own offset, which is what lets the
descriptor keep a minUniformBufferOffsetAlignment-aligned dynamic offset.

red-once (R-16), all three Magma two-process arms, before the backend half:

    DirectVulkan.Split.Buffers.F1WireScenario.SubWordUniformBufferRangePadsMissingBytes
    DirectVulkan.Spawn.Buffers.F1WireScenario.SubWordUniformBufferRangePadsMissingBytes
    DirectVulkan.Tcp.Buffers.F1WireScenario.SubWordUniformBufferRangePadsMissingBytes
        Subprocess aborted - MGPipe: Fatal{UnmigratedVerb, "Magma:uniform-buffer-byte-tail@P7"}

and green after it. The case is the sibling of
ShortUniformBufferRangePadsMissingBytes with a 19-byte range; the byte just past
the range is poisoned, so a rounding that leaks it reads tail.x as 0.5f and the
red channel comes back 127 instead of 0.

Lanes: integration-magma-split 74 (was 73), -spawn 53 (52), -tcp 55 (54),
integration-magma-buffers 20 (19), all green. fatal_census unchanged at 79 abort
sites / 43 family words - these refusals stopped being bare aborts in wave 0, so
a retirement moves no census number and needs no re-baseline. link_ratchet
unchanged at 186.
2026-09-22 16:22:40 -04:00
swung0x48 cf7ca59fab [MG_Remote, docs] (P7): CONTRACT-P7 records package C - vertex-layout split (seven masked reasons), the death-ops debt retired, the opcode-49 white-box deferred to P11, ID-P7-20 2026-09-22 16:09:09 -04:00
swung0x48 074a072739 [docs] (P7): package C's note carries the gate matrix and the three things the integrator has to decide
The per-slice red-once records were written with each slice. This adds what only exists
once the four are together: the gate table (unit 2403, magma split/spawn/tcp 75/54/56 with
the two death-case skips now gone, integration-split 187, integration-spawn 103,
full-split 513 entries 0 red 55 skipped = 458 pass against the 453/57 bar, census
unchanged, parity green, @P7 15/4 -> 13/3, ratchet 186 monotone, G1 .text 0xa52203 with
every section +0 and the pull .so byte-identical), the ratchet bucket split with an honest
note that I did not re-measure the buckets at the branch point, and a numbered list of the
nine things this package changes that live outside its own files.

Three of those nine need an integrator decision rather than a note: §5.4's white-box case
is blocked by R-6 and not by this package; §5.3's g_programResourceCaches deletion is
achieved in the disaggregated build only, because the archive member has to be guarded for
G1; and buffer-window's family word is UnmigratedVerb where ProtocolCorruption is the
honest one, which costs a row in MG_Pipe and an adapter arm that §3.3 deliberately has not
spent.
2026-09-22 16:07:59 -04:00
swung0x48 0acfa9aab9 [MG_State, MG_Backend, MG_Test, docs] (P7): the reflection archive publishes the storage-block index space, and both magma consumers read it
CONTRACT-P7 §5.3, OQ-8. MagmaProgramSource::EnsureStorageBlocks re-ran SPIRV-Reflect over
every stage module of the program PER DRAW: the class is a borrowed view that
WireDraw.inc's WireProgramSource constructs afresh for each wire draw, so its `mutable`
memo was cold every time it was read. The list it rebuilt is a pure function of the linked
program, so it is now built once at link time and travels in the archive.

LinkArtifacts gains storageBlocks{name, binding, dataSize} with its own VisitFields table,
the codec version goes 2 -> 3, and the schema fingerprint and wireFingerprint move with
it - which is expected, and is what makes a mixed-version pair refuse at the handshake
instead of at the first program.

FILLED IN PHASE A, and that is not where it looks like it belongs. The only function that
sees both finished halves is ProgramSpirvTask's L1 cache insert, but its writes reach the
TRANSLATION CACHE only: phase A's artifacts have already been moved into the ProgramObject
by then. A field filled there would be present on a cache hit and absent on every cold
link. So the fill is glslang-derived, and the equivalence is MEASURED rather than assumed:
ProgramTest.TheArchivesStorageBlockOrderIsTheOneSpirvReflectProduces runs the old
consumer's own algorithm over the real modules and compares element for element, on a
multi-stage program with an arrayed SSBO and an atomic-counter block - the three cases
that could have diverged (the instance array collapses to one entry, the synthesized
counter block IS in the backend's space although GL's enumeration excludes it, and the
cross-stage dedupe walks stages in linked order).

MEASURED, same binary and same window (frames=1 draws=1 acc=18, identical bytes),
DirectVulkan x inproc over the SSBO scenarios: spvReflectCreateShaderModule on the wire
draw path 4 -> 0. The 0 is structural: the call site is gone and MagmaProgramSource.h no
longer includes spirv_reflect.h at all, which is the gate a future edit breaks on purpose.

THE MONOLITH CONSUMER IS SWITCHED TOO, under #if symmetry, so the disaggregated build
retires g_programResourceCaches, ProgramResourceCache, StorageBlockResource,
BufferVariableResource, AddBufferVariablesRecursive, NormalizeDescriptorName,
GetProgramResourceCache and the rehash hazard - the open-addressed-map dangling reference
that was a reproducible segfault in ProgramPipelineScenario's two storage-block cases.
Nothing outside DirectVulkan.cpp read any of it (measured; GL_BUFFER_VARIABLE and
GL_ACTIVE_VARIABLES are answered by ProgramInterface.cpp out of a different index space),
so the two arms differ in exactly the two functions their callers use.

THE PULL BUILD KEEPS ALL OF IT, and the integrator should see why: the archive member is
MOBILEGL_BUILD_DISAGGREGATED-only because one more Vector in LinkArtifacts changes its
size, its implicit destructor and its move constructor, and G1 pins the pull build's .text
byte for byte. So §5.3's deletion is achieved in the disaggregated build and not in the
pull one; retiring the pull half is a one-line change the day that pin is retired or the
member moves. Filed as P13 in notes/p7/magma-c.md.

One improvement rides along: the monolith binding getter now asks the program for a
rebind on every read, over an immutable list, where it used to read a binding patched into
a mutable cache - so a rebind can no longer be lost to a cache rebuild that raced a
state-version bump.

G1: .text 0xa52203 unchanged, 27837 -> 27837 defined symbols, 0 added / 0 removed / 0
resized / 0 renamed, file byte-identical. An unguarded call site caught the guard rule
once mid-slice by failing the pull compile, which is the loud version.

RED-ONCE: publishing nothing reds the order-equivalence case (with the reference list
printed) and 25 of 60 storage-block scenarios - the MONOLITH DirectVulkan arm beside all
three transports, which is the evidence that the monolith consumer really was switched.
Restored: 61/61.
2026-09-22 16:07:59 -04:00
swung0x48 f1b04edfc6 [MG_Backend, MG_IntegrationTest, docs] (P7): the magma vertex-layout refusal splits into one protocol fatal and seven masked attributes
CONTRACT-P7 §3.2's `vertex-layout` and `vertex-format-conversion` rows.
BuildWireVertexInput answered eight different conditions with one `false`, and the single
caller answered that with a P7-marked MagmaWireFatal. So a Magma split session that met a
GL_FIXED array, a dvec on a device with native fp64, or any format without
VK_FORMAT_FEATURE_VERTEX_BUFFER_BIT did not decline the draw: it killed the process. The
monolith arm on the same device has always masked the attribute out and then declined the
draw loudly and once.

The split is by what the reason is ABOUT, not by severity. `buffer-window` is a statement
about the RECORD - an enabled attribute the program reads whose slot lies outside the
window set_vertex_buffers published - so it stays a named Fatal through the hook; it is
what caught the Redmi VAO hash collision fixed in 376c04be. Every other reason is a
statement about what Vulkan can express on this device, and those now set
unsupportedAttribMask and continue, with the monolith arm's own pre-flight in WireDraw.inc
declining the draw when a masked bit is one the program actually reads.

The contract says "the other six"; there are seven. `offset-overflow` post-dates the
audit's count and is an arithmetic guard rather than a shape one. It is masked with the
rest: an attribute whose base offsets cannot be added is one Vulkan cannot fetch, the
monolith arm never computes that sum at all, and masking is strictly safer than a Fatal
for a number no application can reach on purpose.

unsupportedAttribMask now enters the wire layout hash, for the reason the monolith entry's
hash already carries it: two VAOs can otherwise agree on every binding, attribute and
divisor and differ only in which enabled attribute was masked, and a blind hash would hand
one of them the other's pipeline. Omitting it was safe only while the mask was always zero.

vertex-format-conversion becomes a decline for a narrower reason: the converter is only
ever asked for a conversion BuildWireVertexInput already chose, so a refusal means the two
disagree - a defect on this side, not a peer's malformed record.

@P7 census: 15 sites / 4 files -> 13 / 3. Renderer/WireDraw.inc leaves the list entirely.
fatal_census.py unchanged at 79 abort sites / 43 family words / 0 unmarked, as expected:
wave 0 already funnelled these, so retiring a MagmaWireFatal CALL moves no abort count.

New MagmaVertexLayoutScenario, DirectVulkan only (Espryt hands GL_FIXED to a driver that
supports it, so the same name would assert a different fact there). DataType::Fixed32 has
no arm in ToVkVertexFormat at all, which makes it the one unmappable shape reachable on
host lavapipe - §2.5 records that no other P7-marked refusal is. One case has the program
READ the array and asserts the draw declines; the other leaves it enabled but unread and
asserts the draw lands, which is what stops the first being satisfied by "decline every
draw that mentions GL_FIXED". Registered on all three transports and picked up by the
ambient monolith DirectVulkan discovery, so the A/B is direct.

RED-ONCE: `format-map` put back on the protocol verdict reds the three split arms with
`Subprocess aborted` / `Fatal{UnmigratedVerb, "Magma:vertex-layout-buffer-window"}` while
the monolith control stays green - exactly the divergence this closes. Restored: 10/10.

The gating Magma lanes grow with the new entries: split 73 -> 75, spawn 52 -> 54, tcp
54 -> 56. Growth only.

Open for the integrator: the honest family word for buffer-window is ProtocolCorruption,
and MGPipeFatalFamily carries only UnmigratedVerb and RoleViolation. Adding a third costs
a row there plus an adapter arm, and §3.3 keeps the three Magma funnels on UnmigratedVerb
deliberately so the refusal census stays comparable to P5b. Flagged, not decided.
2026-09-22 16:07:59 -04:00
swung0x48 1255caa2fb [MG_Backend, MG_Impl, MG_Util, MG_Test, MG_IntegrationTest, docs] (P7): the server publishes kCapResidentSubData off its own wire resource table
CONTRACT-P7 §5.4, OQ-10. InitServerRoleCommon computes the whole capability mask and
never ORed this bit, so MGPipeResourceOpsHaveSubDataResident answered false under every
transport and BOTH backends were pinned to the ordered in-place host memcpy - which made
Magma's SubDataResident arm (VkBufferManager.cpp's g_vulkanWireResourceOps) dead code on
the wire even though the arm has been there since P5f.

The bit is derived from g_resourceOpsAtStep2->SubDataResident, i.e. from the pointer this
function already pinned at step 2 and that step 5 already refuses to see swapped - never
from ActiveBackendType. That is ID-39's lesson and not a style preference: a mask is a
statement about THIS backend's registered table, and a bit derived from the enum would
keep claiming the capability for a build, a flavour or a future backend whose table does
not carry the arm.

`rsd=` joins the emit[] bracket on the PipeStats summary line, counted at the emitter once
per buffer_subdata_resident record rather than once per call, because a write past the
content chunk cap is cut into several records. It is the only observable that separates
the resident arm from the memcpy beside it: the bytes in the buffer are identical either
way, which is exactly how the capability stayed unpublished for a phase with every lane
green.

WHAT THIS DOES NOT YET PROVE, and R-6 is the reason rather than an omission. §5.4 asks for
a white-box case showing opcode 49 on the wire. None can exist at this tree: the resident
arm is reached only from a GPU-RESIDENT store, residency comes only from
AdoptPersistentMap, and every persistent-map acquisition declines under split by R-6
(PipeApply.cpp:2331 "A SPLIT BUILD RUNS AT TIER T2 AND DECLINES EVERY ACQUISITION,
ALWAYS"; T0/T1 are P11's and AdoptTierIsEmulate is a named Fatal for them). So the
capability is wired AHEAD of its consumer, deliberately, and `rsd=` is what will show the
records the day P11 lands a tier that adopts. Filed for the integrator in
notes/p7/magma-c.md.

What IS proved is both halves of the bit itself. RemoteClientTest's Magma bootstrap case
now asserts the published bit EQUALS `ops->SubDataResident != nullptr` - the two compared
to each other rather than both to true, so the case also covers a backend without the arm
on the day one exists - and a new CtWireScenario case reads the bit out of the client's
own caps mirror, through the same accessor the frontend reads, on both backends and all
three transports.

RED-ONCE, both levels. Deleting the OR reds
CapsMirrorTest.MagmaTransportPublishesRealBufferConsumersWithoutRunAhead
(RemoteClientTest.cpp:692, "Which is: false / Which is: true") and all SIX arms of
DirectGLES/DirectVulkan x {Split,Spawn,Tcp} Ct.TheServerPublishesTheResidentSubData
CapabilityFromItsOwnTable ("the client's caps mirror did not receive
kCapResidentSubData"). Restored: green.

Also carries the P4a death-helper comment from the previous commit's sweep: it no longer
says Magma installs no death table, because it now does and still frees no slot.
2026-09-22 16:07:58 -04:00
swung0x48 dce9953d36 [MG_Backend, MG_State, MG_Pipe, MG_Remote, MG_Impl, MG_IntegrationTest, docs] (P7): magma installs a state-object death table, and the inproc arm stops losing every object_death
CONTRACT-P7 §5.5. DirectVulkan had no StateObjectDeathOps at all, so under a Magma
inproc session NotifyStateObjectDestroyed reached nobody and the server was never told
that a frontend object had died. Package L measured the shape: with the WireTables
install condition made backend-agnostic, CtWireScenario's two death cases PASS on Magma
spawn and tcp - the remote client has no local backend, so the client-side emitter
installed there answers the notice - and FAIL on inproc, where the install condition
deliberately stands aside for "the backend's own dispatcher" and Magma had none.

g_magmaStateObjectDeathOps is that dispatcher. It mirrors g_glesStateObjectDeathOps
(DirectGLES/Managers.cpp:335) with ONE arm rather than two: Espryt's other arm destroys
a backend twin keyed by lifetime id, and Magma keeps no such twin - its two
client-minted kinds are keyed on {slot, gen} by MagmaPipeIdentityTables and the other
six are still reached from their frontend objects - so what is left is exactly the emit
of object_death off the apply thread. Installed from
BackendObject_DirectVulkan::Initialize(), i.e. ServerLoop::CreateBackend, step 1 of
InitServerRoleCommon: the one place BOTH Magma server roles pass through, and before any
client session exists. #if MOBILEGL_BUILD_DISAGGREGATED && MOBILEGL_PIPE_PUSH, so the
pull build compiles nothing new.

The two skip guards in CtWireScenario go with it. Every other "Magma installs none" site
the audit found is comment text whose ARGUMENT survives, and each is reworded rather than
deleted: the table emits the wire record and frees no client slot, so P4a's
backend-neutral NotifyAndFree is still the only thing that returns one, HandleRecycle's
C-1 case still measures the only protection the allocator has, and MagmaPipeArms' age
sweep is still the only thing that retires an identity slot.

The framebuffer death case now reads the client's allocator beside the server's tally.
That correction matters: the plan called this a slot leak provable with PipeSlotPeek, and
it is not - it has been backend-neutral since P4a. What leaked was the SERVER's twin, and
a framebuffer has no wire delete opcode at all, so object_death is its only delivery.

Two installers cannot race although the global is last-writer-wins with no stacking:
Magma's runs only in a process that owns a DirectVulkan backend, and
InstallClientWireTables' arm runs only under Transport == Spawn, which is by construction
the process that owns none.

RED-ONCE: `if (false) InstallStateObjectDeathOps();` reds the two Split cases alone
(`Expected: (afterCounters.deaths) > (deathsBefore), actual: 0 vs 0`, CtWireScenario.cpp:264)
and leaves spawn and tcp green - the exact split of responsibility the comment claims.
Restored: 8/8. Recorded in docs/Disaggregated/notes/p7/magma-c.md.
2026-09-22 16:07:13 -04:00
swung0x48 07079c3c0e [docs] (P7): X1 verified on the phone, F1 verified but acquitted, OpenRA is a cold-cache timing race (ID-P7-18/19)
X1: 26.3 and create-instancing pass under inproc on both backends, zero declined
faults, tracker line confirms tag=0xb4. F1: the R-8 line is gone in every run, yet
OpenRA reads 0.976494 on a cold driver cache (three pm-clear runs identical), 1.0
warm, 1.0 cold monolith, and FRAMESINFLIGHT=1 flips it half the time - a GPU-timing
race on the wire readback/present path, handed to package B as slice 0.
2026-09-22 16:05:39 -04:00
swung0x48 3179c4972d [docs] (P7): F1's emitter, its three red-once arms and the one thing lavapipe cannot prove are on record
`notes/p7/f1-caps-placeholder.md`, per CONTRACT-P7 §9's "one red-once per package
recorded in notes/p7/". It carries the call chain from `MobileGL::Initialize` down
to `CapsMirror::ServerConsumes(0x400)`, the three corrections to the brief's reading
(it is not a race; P6.5's fix already covers inproc; a bounded wait alone deadlocks),
the three red-once arms, and the two things still owed to the integrator.

The arm worth reading twice is A. On lavapipe the BASE build carries the R-8 line on
every transport, with and without the delay knob, and OpenRA still scores
ssim 1.000000 / mismatch 0. So the eleven dropped resource_creates cost nothing
visible there, F1's red-once is the LOG LINE and the new Fatal rather than an SSIM,
and this package must not be read as having explained the device's 0.976494 /
0.988998. That still needs the phone.
2026-09-22 15:55:08 -04:00
swung0x48 22c7361c5f [MG_Test] (P7): the placeholder caps cases assert the refusal and the wait instead of the answer
`CapsMirrorTest.APlaceholderMirrorConsumesNothing` asserted the defect F1 closes -
"with no snapshot the mask is zero, every family answers no consumer, the client
emits nothing and the legacy pull path runs" - and it was green for five packages
because nothing on the client ever checked whether that pull path exists. It does
not. Four cases replace it, and each one is the control for a different way the
new gate could go wrong:

  * `APlaceholderMirrorRefusesToAnswerAndNamesTheFamily` forks, asks a placeholder
    mirror about TextureResources with no session anywhere, and requires the death
    to carry `Fatal{CapsBeforeFirstSnapshot, "TextureResources"}` AND `0x400` -
    R-16's rule, because "something aborted" is satisfied by any of the other 43
    family words.
  * `ThePlaceholderWaitsForASnapshotAnotherThreadIsFetching` drives the PRODUCTION
    bring-up (`MG_Backend::Init`) with `MOBILEGL_TEST_DELAY_FIRST_CAPS_MS=50` and
    asks from a second thread. The wait must return the server's real answer, not
    the placeholder's false and not a death. It is the only shape in which "block
    with a bounded wait" is a correct answer to F1.
  * `ARealSnapshotThatWithholdsAFamilyStillAnswersFalse` is what keeps this from
    being a callMask defaulted to all-ones: a server that really consumes nothing
    is still answered "no", counted and named. Only the PLACEHOLDER is refused.
  * `TheReadAccessorsStillAnswerFromThePlaceholder` pins the half of P5's ruling
    that F1 does NOT narrow.

`ResetCapsMirrorForTest()` is new and test-only: a fork child inherits whatever the
previous case adopted, so without it a case about the placeholder would be a case
about that snapshot.

`ServerLoopTest`'s main gains no code, only the sentence that explains why it has
none: it sets `MG_Config::Transport` by hand and never runs `MG_ConfigLoader::Init`,
which is exactly what keeps the client-side gate disarmed for a binary that builds
server sessions and the server's own frontend GLContext and never a client at all.
2026-09-22 15:55:08 -04:00
swung0x48 df17c416c0 [MobileGL, MG_Backend, MG_Remote, scripts] (P7): the split wire comes up before the first frontend object, and the caps mirror stops deciding a family's fate from a placeholder
F1. `MG_State::Init()` runs before `MG_Backend::Init()`, and the second is what
starts the session - so `GLContext`'s `TextureState` constructor materialised its
eleven default texture objects, each of them emitted `resource_create` through
`FamilyIsLive`, and `CapsMirror::ServerConsumes(0x400)` answered every one of them
from a generation-0 placeholder with `callMask=0`. All eleven records were dropped
and the whole event was one WARN line:

  MG_Remote client: the server does not consume MGPipe subsystem 0x400 - this
  family emits NOTHING and the legacy pull path runs for it (R-8). callMask=0x0,
  caps generation 0

It is NOT a race. The line reproduces byte-for-byte on lavapipe and on the Adreno
830, because the two Init calls are two statements on one thread. And "the legacy
pull path runs for it" is false on the client under split: there is no pull path
on this side of the wire, so withholding is not a fallback, it is a drop that only
a later respecify's self-healing create ever undoes.

Two of the three options the brief offered do not apply, and both facts are worth
recording. (b) "extend the P6.5 fix to inproc" is already the tree: `FinishStartup`
step 7 is the one funnel every transport reaches and it will not return without a
real snapshot. (a) "a bounded wait in Consumes()" cannot work alone and would
deadlock: at `MG_State::Init()` no handshake is in flight, and the one that will
land is started later BY THE SAME THREAD.

So the fix is the ORDER, with the rule behind it as the gate:

  * `MobileGL::Initialize` calls `MG_Backend::InitSplitRolesBeforeState()` before
    `MG_State::Init()`. `Init()`'s split arm becomes one body, `InitSplitBackend()`,
    shared with the direct callers that still reach `Init()`; the monolith arm keeps
    its place after `MG_State::Init()`, because moving it would move pull `.text`.
  * `CapsMirror::ServerConsumes()` refuses to answer from a placeholder. It waits,
    bounded, when a bring-up is genuinely in flight on ANOTHER thread, and otherwise
    dies as `Fatal{CapsBeforeFirstSnapshot, "<family>"}` through `SessionFail` -
    a new family word, one row in `FatalFamilies.def`, no new abort site.
  * Two arms are exempt and say why in place: the server arm (`MGPipeServerArm()` -
    the apply thread and the spawn server, where the client's mirror is the wrong
    object rather than a late one), and a process whose own configuration never
    asked for a split transport (`MG_Config::SplitTransportRequestedByConfig`, set
    by `MG_ConfigLoader::InitTransport`, which is what tells a real run apart from a
    fixture that assigns `MG_Config::Transport` by hand).

The read accessors keep P5's ruling untouched: `LogBackendInfo` reads
`GetRendererInfo()` during `MG_Backend::Init()` and an abort there is a process that
cannot start. Only the DECISION changed. The warning was not made quieter and the
callMask was not defaulted to all-ones.

`MOBILEGL_TEST_DELAY_FIRST_CAPS_MS` (server side, disaggregated builds, inert unless
set) holds back the FIRST CapsSnapshot. On spawn/tcp that is a genuinely late
snapshot in another process and it exercises the client's step-7 wait; on inproc it
only makes `Start` slower, which is itself the proof that the only window a client
can read a placeholder in is the one before `Start` is called.

G1: `.text` 0xa52203 and the 30570-symbol list identical to p7-before/pull-syms.txt.
Census: 79 abort sites unchanged, 44 family words (was 43). Link ratchet: 186,
unchanged.
2026-09-22 15:55:08 -04:00
swung0x48 206abdfa44 [docs] (P7): the gate-5 CTS base reading is on record - monolith x DirectVulkan x KHR-GL46 on the Redmi, five complete blocks
shader-image 63/56/6, ssbo 124/0/0, dsa 368/0/2, texture 815/221/18, packed-pixels
4732/0/0 (Pass/NotSupported/Fail); every block complete (no unrun, hung or crashed);
library p7w1-d260f110, glcts from VK-GL-CTS 0b04c470e with the FBO colour-texture
patch, API probe confirms GL 4.6. The AFTER reading compares inproc against this at
0.5 pp per block with new crashes as a separate hard red.
2026-09-22 15:33:52 -04:00
swung0x48 5c4235d7ba [docs, MG_Remote] (P7): ID-P7-12..17 - gate 3 per case against monolith, photon excluded, wave-0 naming ratified, X1 on the critical path, gate 5 is five blocks at GL46, E2/E0b folded into the F1 re-test 2026-09-22 15:25:20 -04:00
swung0x48 ec3ba00af3 [tools] (P7): the CTS skip list the runbook already passes exists
run_cts.py silently tolerates a missing --skip-file, so the known device-killers from
the skill (clip_distance.functional, the multisampled_to_singlesampled blit family,
the rgb565 repeat-mode case) would each have cost a reboot in the gate-5 window.
2026-09-22 15:16:37 -04:00
swung0x48 4cbf78c269 [MG_Remote, MG_Test, docs] (P7): the persistent-map tracker answers faults in the kernel's address space, and a map hands back a range it can write
Wave 2 package X1. Three CI split-subset traces died on the Redmi under inproc and
passed under monolith on the same device and APK, on the client thread, in the
replayer's memcpy into a pointer MobileGL had handed it: SIGSEGV / SEGV_ACCERR on the
destination's FIRST byte, destination page-aligned.

The tombstone names the same address twice, in two different address spaces:

    x0  b4000073ac9a5000 ... x2 0000000001800000
    signal 11 (SIGSEGV), code 2 (SEGV_ACCERR), fault addr 0x00000073ac9a5000

x0 is the pointer the application writes through, carrying bionic's 0xb4 heap tag
(the same tombstone reports tagged_addr_ctrl PR_TAGGED_ADDR_ENABLE). fault addr is
si_addr, which the kernel reports UNTAGGED.

TrackWriteMap took shadowBase from MappedData() - a tagged heap pointer - kept the tag
through the page masking, mprotect()ed the mapped range PROT_READ and published the
TAGGED base/end into the slot. The handler compared si_addr against it:

    if (base <= kSlotSettingUp || address < base) continue;

0x73ac9a5000 < 0xb4000073ac9a5000 is true for every fault, forever. Ownership test (2)
never matched, and every fault on a page the tracker had itself protected was chained
to debuggerd. mprotect was fine throughout - do_mprotect_pkey untags its own argument -
so the right pages really were read-only; only the recognition was broken. x86-64 has
no top-byte tag, which is why every host lane has always been green.

The fix is two rows and one ordering:

  * UntagAddress(), applied to the shadow base at registration, to the shadow base in
    PushBlocksFor (so trackedBase - shadowBase stays in one space) and to si_addr in
    the handler. Written WITHOUT an #if defined(__aarch64__): masking bits 56-63 is
    the identity for every valid userspace address off arm64, and an arch guard would
    have put the only gate that can catch this on the one platform CI never runs.
    Dereferences keep the tagged pointer they came from.
  * SlotOwnsFault(), the ownership row the handler and the unit case both read.
  * REGISTRATION ARMS NOTHING. AcquireMemoryRange registered and only then returned
    Bytes() + range.start, so glMapBufferRange handed the application a PROT_READ
    pointer whose writability depended on a handler reaching the writing thread - and
    the application is free to write a mapped arena from a worker that blocks every
    signal. Nothing is lost by waiting: every page bit is already set at registration,
    so the first push ships the whole interior regardless, and it is that push which
    arms the pages. 6144 pointless faults per 24 MB arena go with it, and the
    invariant the file rests on (bit set <=> page writable) is now true from
    registration onward instead of only from the first push.

Diagnostic, independent of the fix: one MGLOG_I_ONCE at the first registration naming
the mapping the way the device run needed and did not have (allocator, pointer, ITS
TAG, extent, range, published span, pages, alignment arm), and a decline record in the
handler - relaxed atomic stores plus one async-signal-safe write(2) - readable through
DeclinedFaults(). A non-zero decline count whose si_addr lies inside nearest-tracked
once the top byte is masked off is this defect's signature and nothing else's.

Not the mechanism, and checked: libsigchain (the tombstone's "F libc : Fatal signal 11"
is bionic's own handler, so the chain ran), the 16 KiB-page refusal (the device is a
4 KiB kernel and the protection demonstrably took), RefaultOfANonWriteAccess (matched
is false long before it is consulted), and the P5d shadow allocator (correct - it is
what makes the destination page-aligned in the tombstone).

Red-once, two and orthogonal (MG_Test/Wire/PersistentMapFaultTest.cpp, 6 cases):
dropping the two UntagAddress calls in SlotOwnsFault reds only
AKernelsUntaggedFaultAddressIsOwnedByATaggedShadowsSlot, which runs the device's own
two addresses through the real predicate on an x86-64 host; restoring the mprotect at
the end of TrackWriteMap reds only AMapHandsBackARangeWritableWithoutTheFaultHandler
(the child dies by SIGSEGV) and RegistrationLeavesEveryPageOfTheRangeReadWrite.
AnArmedPageIsStillAnsweredByTheHandlerAndMadeWritable is the anti-vacuity case.

Gates: ctest -L unit 2408/2408; integration-split 186/186; integration-spawn 102/102;
integration-magma-split 73/73 (PersistentCoherentMapScenario green in the Split, Spawn,
PersistentMapArm and SmallRing variants); fatal_census.py 0 unmarked. G1 holds - pull
.text 10822147 = 0xa52203 unchanged, symbol report 0 added / 0 removed / 0 resized /
0 renamed, nm --defined-only identical at 30570 and nm -D at 11933. Every line is under
MOBILEGL_BUILD_DISAGGREGATED (CMakeLists.txt:592-659), so the pull build sees none of it.

Evidence and the device experiment: docs/Disaggregated/notes/p7/x1-persistent-map-segv.md
2026-09-22 15:16:37 -04:00
swung0x48 4f2d4c4e83 [MG_Remote, docs] (P7): CONTRACT-P7 carries wave 0's numbers, and device window #1a is on record
CONTRACT-P7: the lane counts (magma-split 73 / spawn 52 / tcp 54; full-suite 510 x 3
with one StreamLink x Magma red), the census baseline (79 abort sites over 20 files,
43 family words, 0 unmarked), the ratchet baseline (186 = p7-magma 101 / espryt 14 /
both 60 / core 11; espryt's clearable count is 9) and the @P7 count (15 sites / 4
files) replace the 待填 markers.

Device window #1a (Redmi 2f7cbe2e, APK p7w1-d260f110, reboot-clean): E0a monolith
39/39 deterministic, three below threshold on monolith itself (create-indirect,
1.21.11 main-menu, iris-photon - Adreno refuses its pipelines with VK_ERROR_UNKNOWN);
E0a inproc: OpenRA 0.976494 (run-ahead) / E1 lockstep 0.988998 - both historical
numbers were OpenRA, and its client log shows a TextureResources verb answered from
caps generation 0 (package F1); 26.3 / create-* SIGSEGV in the replayer's memcpy on
BOTH backends and with the persistent-map push off (package X1); iterationt x2 die
by name on Magma:mipmap-shader-format-or-shape (package B); bsl-esc-menu hits
scudo OOM under inproc. Gate 3's denominator is 36; run-ahead and present credit
are off the suspect list.
2026-09-22 14:49:42 -04:00
swung0x48 801f8edaa6 [scripts] (P7): the parity self-test patches the seam the Magma tiers read through
Package D added a test that patches lane_cases and hands spawn_lane_parity.py a
placeholder build dir; package L then made the three Magma tiers read through a new
lane_names, so the placeholder reached a real ctest and the unit label went red on the
integrated tree. Patching lane_names to an empty set on every arm makes those tiers
report "nothing to compare" and leaves the assertion about lane_cases intact.
2026-09-22 14:43:51 -04:00
swung0x48 8bb6309a1c [scripts, CI, docs] (P7): the 184-symbol link-closure ratchet lands as a gate
CONTRACT-P6.md 12.1 has carried "a ratchet recomputes the 184 in CI and asserts it only
ever falls" as an obligation since P6. The roadmap handed it to P6.5; the P6.5 line never
received it, and scripts/ci/ has never had anything that computes this number. ID-P7-7
gives it to P7 wave 0. This is it.

scripts/link_ratchet.py mechanises a6's hand experiment: partition the per-object output
of the single MobileGL target into SERVER / FRONTEND / SHARED, then compute
(SERVER undefined) - (SERVER defined) - (SHARED defined) intersect (FRONTEND defined),
which is the undefined-symbol list a real link of a server-only image would print. The
partition is an explicit table with a rationale per row, because it IS the module boundary
the build system does not have (SOURCE_FILES is one flat list; boundaries are P13's work).
An object matching no row is a hard failure, as is a build whose CMakeCache lacks the three
disaggregated options, or an empty role set - every way this gate could go green without
having measured anything is an explicit exit 2.

Measured on 78b7d6be: 186 symbols. The three backend buckets reproduce a6's hand-made
101 / 14 / 60 exactly, as do its per-object referrer and definer distributions
(UniformManager 90, VulkanRenderer 60, Managers 53, DirectGLES 51; TextureObject 59,
SamplerObject 30, BufferObject 18, FramebufferObject 16, PipeFill 9). Nothing from a6's
184 has been cleared; the +2 are both P6.5's:
  Client::ClientSession::StartSpawned()                 - fork-per-session
  GLState::ProgramArtifactsSchemaFingerprint()          - archive v2 caps

The baseline stores the LIST, not a count (ID-P7-7): a count-only ratchet goes green on the
day one symbol is cleared and another added. --assert-monotone fails only on symbols absent
from the baseline and names each one's defining object and every referring object; symbols
that disappeared are reported with an instruction to re-baseline in the same commit, and
never fail.

The CI step goes in build-linux-split, not build-linux: build-linux takes the three options
from MOBILEGL_CI_DISAGGREGATED, which is ON only on feat/disaggregated and
codex/fix-github-ci, so on any other branch it builds a monolith and the gate would measure
a SERVER set with no transport half. The JSON is uploaded even when the gate fires.

Two findings for the integrator, both written up in the note:
  - a6's four published bucket totals sum to 181 against its own headline of 184. These
    buckets are a strict partition and always sum to the headline; the difference is in
    a6's core bucket, which it listed selectively.
  - 5 of the 14 "P3b/P4b" symbols are MG_Pipe slot-allocator and apply-thread plumbing
    that sits under MG_Impl/Pipe/ by file location alone. P3b/P4b cannot clear them by
    migrating Espryt functionality; they need P13's module move. That bucket's floor is
    5, not 0, so the plan's 0.5-day Tier 1 estimate for "D12's 14 symbols" should be
    re-cut against 9. All 7 symbols from that object carry a `# P13` annotation.

red-once (R-16) is recorded verbatim in the note. Worth knowing: the plan's own example
(SamplerObject::GetMinFilter) does NOT turn this gate red, because it is already in the
baseline - the ratchet reddens on new symbols, so a negative control has to pick one that
is not in the 186. A reach to GetBorderColorForm() from DirectGLES/Utils.cpp does, naming
the symbol, its definer and its referrer, exit 1; reverted and green again.
2026-09-22 14:26:04 -04:00
swung0x48 1de8f02ebe [MG_IntegrationTest, docs] (P7): the whole integration binary runs under Magma split, and it is one red
Tier 2 replays the split-arm registrations - ~100 scenarios, the set Espryt's disaggregation work
has carried since P5 - and it measured green on Magma across all three transports. That is a real
result but it is NOT exit gate 1. The gate says the INTEGRATION BINARY is green under Magma split,
and the set that names is the one the monolith DirectVulkan. arm runs: the whole suite minus
MGL_ITEST_MONOLITH_MECHANISM_FILTER. 510 cases against tier 2's 102, and the readback matrices,
the same-texture copy shapes and the odd-offset uniform ranges are in it and nowhere else in the
disaggregated registry.

integration-magma-full-{split,spawn,tcp}, in no gating step, private log per entry through a list
of its own. THE RESULT: 453/0/57 on inproc, 453/0/57 on spawn, 454/1/57 on tcp - and not one of
the 1530 entries reaches any of the tree's fifteen @P7 named refusals. The single red is
IterationRPProgram203Scenario.FixedCompleteInputProducesFixedCompleteGoldenOutput on the TCP arm:
one texel of 262656, texel 0, which is the case's own exposure value
(IterationRPProgram203Scenario.cpp:371, actual half bits (0x357a, 0x4b65) against golden
(0x3a74, 0x4821)). Deterministic 5/5 on tcp; 5/5 GREEN on inproc, on spawn and on the monolith
control. spawn and tcp differ only in MOBILEGL_IPC_DATA, so the suspicion is the STREAM data path
and not "two processes"; the GLES tcp lane is green in the same run, so it is not the fixture.
Not fixed here - it is what the census is for.

TWO ENVIRONMENT DIFFERENCES FROM THE GATED TIERS, both measured rather than assumed:

  * NO MGITEST_SPLIT_LANE=1. That marker arms ScenarioFixture's destructor assertion - a case in
    an armed split lane must have moved the client encoder's record ordinal - which is right for
    a curated lane and wrong for a whole-binary census. With it set, the first run produced 13
    reds, all `EmitSeq was 20 at SetUp and is 20 now`: AdvertisedLimitsScenario (8),
    SpirvShaderBinaryScenario (4) and ProgramPipelineScenario.AReservedName - a limits query is
    answered from the client's CapsMirror and a shader-binary error surface never reaches a verb.
    Those 13 would red on a DirectGLES census too. What it gives up is stated in the comment: this
    tier does not prove its records crossed, and the tcp topology proof is inert in it.
  * DualBlockScenario.* filtered out. Its green REQUIRES MOBILEGL_IPC_ROLE_SPLIT_STATE=1 and a
    census must not set that knob; without it the case is red by name, which is its documented
    negative-control behaviour. The monolith arm skips it for the same class of reason.

CtWireScenario gets MOBILEGL_TEST_SERVER_COUNTERS=1 in this tier - a switch that makes a counter
readable, not an oracle that asserts - because without it the spawn arm's ApplierReset case reds
on "server control-record telemetry missing", which says nothing about Magma. The strict-arming,
pmap and persistent-map-arm markers stay out.

The inproc-only exclusion in spawn_lane_parity.py is now PER TIER: in the gated tier the two
scenarios are registered on the inproc arm alone, in the census they are on every arm and simply
skip, so excluding them there would manufacture a spurious "extra".

Measured after the change: every pre-existing lane unchanged and green (magma-split 73,
spawn 52, tcp 54, all-* 102/102/104, GLES 186/102/104, buffers 19, runahead 14, caches 7), the
dual-block census 257 green / 2 skipped / 0 red with an empty ratchet, both new CI steps green,
parity RC=0 across all three tiers, G14 4155 -> 6095 with removed=0.

The note now carries the grouped red list, the 57 named skips by reason, the four pieces of
evidence that the tcp red is a defect and not plumbing, and six findings - the first being that
wave 2's clusters cannot be scheduled from this lane's red list, because on lavapipe there is
essentially none.
2026-09-22 14:24:19 -04:00
swung0x48 78e10ccf80 [CI] (P7): retrace-split exports every Magma knob the monolith retrace job exports
Integrator ruling, completing the alignment package L started. The split retrace job now also
exports MOBILEGL_MAGMA_R11G11B10F_FALLBACK for every DirectVulkan row and
MOBILEGL_MAGMA_DISABLE_BLENDED_DEPTH_WRITE for improved-transparency-minecraft-26.3, copied
from the monolith job (:2180 / :2195) with the same guards.

The point is comparability, not the knobs: a split retrace whose monolith control ran a
different repair set cannot say whether a divergence is a transport fact. With the three
iterationRP repairs from the previous commit, every DirectVulkan row of the 154-cell split
matrix now runs the quirk configuration its control runs.

retrace-verify is left alone, as the ruling says.
2026-09-22 14:24:19 -04:00
swung0x48 a7425d95e0 [MG_IntegrationTest] (P7): every Magma lane pins the ICD the monolith arm has always pinned
Integrator ruling on the asymmetry package L reported. The six hand-laid Magma lanes (Fm,
Buffers, RunAhead, Caches, NamedBlit, P5fRsp) appended ${MGL_ITEST_COMMON_ENV}, so the monolith
DirectVulkan. arm ran with VK_ICD_FILENAMES pinned and the three lavapipe iterationRP repairs
forced on, while the 73 integration-magma-split entries ran against whatever ICD the loader
found first with those repairs at their device auto-detection default.

That is a latent flake rather than a semantic: the day a runner or a workstation grows a second
ICD, the lane measures a different driver from the gate it mirrors and nothing in its output
says so. Consistency wins - all six now append ${MGL_ITEST_VULKAN_ENV}, the same list the new
MGL_ITEST_VULKAN_{SPLIT,SPAWN,TCP}_ENVIRONMENT use.

The ICD pin follows the BACKEND, not the lane, in the two blocks that register both: NamedBlit
and P5fRsp loop over DirectGLES and DirectVulkan, and handing VK_ICD_FILENAMES to the DirectGLES
half would pin an ICD for a lane that never opens Vulkan. Verified in the discovery: all 73
integration-magma-split entries carry VK_ICD_FILENAMES and all three repair variables, and the
two DirectGLES.Split.NamedBlit. entries carry neither.

Re-run after the change, as the ruling asks: integration-magma-split 71 PASS / 0 FAIL / 2 SKIP
out of 73, unchanged - the two skips are the same ClipDistance pair with the same reason.
2026-09-22 14:24:19 -04:00
swung0x48 faca09d4ad [CI, docs] (P7): retrace-split exports the iterationrp knobs, and the two new Magma lanes gate
Three changes, one theme - making the two halves comparable, then gating what measured green.

retrace-split's "Retrace and validate under MOBILEGL_TRANSPORT" step exported no Magma knob at
all, while the monolith retrace job (:2183) exports three iterationRP repairs for
minecraft-1.21.4-fabric-iris-iterationrp-in-world x DirectVulkan. Those repairs are tri-state
quirks defaulting to device auto-detection and lavapipe is on no auto list, so that cell of the
split matrix (154 cells, and it contains that case x DirectVulkan x {inproc, spawn}) was being
compared against a monolith control running a different repair set. The same backend+case guard
is now copied verbatim into the split job.

Two gating steps for the lanes this wave added, and ONLY because they measured green on the
first run: integration-magma-{spawn,tcp} run like the existing magma-split step, and the
nineteen application-buffer cases get require_green BY NAME on both new arms - the tcp arm's
required list includes TcpServer.Start / Stop on purpose, because ctest lists the fixture
entries under the label and a lane whose supervisor failed to start has nothing to say about the
nineteen. The informational integration-magma-all-* tier is deliberately in no gating step.

Both steps were executed locally, not merely linted: spawn 19/19 and tcp 19/19 with
"require-green: N PASS, 0 skip, 0 failed, complete discovery". All 99 run: blocks in the
workflow pass bash -n.

The note records the first run in full: per-lane PASS/FAIL/SKIP, the two blocker groups the
first registration attempt produced (both lane plumbing, neither a server abort, wrong pixels
nor a hang), the arming evidence, both red-once demonstrations, the parity and G14 numbers,
what was NOT verified (no device, no retrace run, no G1) and six findings for the integrator -
the first of which is that P7 exit gate 1 has no host-side red list at all, which the wave 2
plan is built on top of.
2026-09-22 14:24:19 -04:00
swung0x48 e7e3265c47 [MG_Remote] (P7): the wire client's death notice is installed for any backend, not just Espryt
WireTables.cpp:740 installed the client-side object_death emitter only when
ActiveBackendType == DirectGLES. P6.5 wrote that because DirectGLES was the only backend with a
two-process lane, so the condition read as "the case this can happen in" rather than as a policy
- and under a Magma spawn or tcp client it is a silent hole of exactly the shape P6.5 closed for
Espryt: no notice installed, NotifyAndFree drops every object_death before freeing the slot, and
CtWireScenario's two death cases report ObjectDeaths=0 and prove nothing.

The two remaining conditions carry P6.5's intent, both halves of it: Transport == Spawn (this is
the remote client, the one with no local twin table; under inproc the backend's own dispatcher
is the right one) and GetStateObjectDeathOps() == nullptr (whoever installed first keeps the
notice).

RED-ONCE: with the install site short-circuited, both DirectGLES spawn death cases red on
CtWireScenario.cpp:187/:246 with

    Expected: (afterCounters.deaths) > (deathsBefore), actual: 0 vs 0

which is the silent zero this closes. Restored and green.

WHAT IT DOES NOT DO is give Magma a death table - that is wave 2 package C. But it already does
more than the plan predicted: with CtWireScenario's own BackendName() != "DirectGLES" skip guard
temporarily lifted (a probe, not committed), both DirectVulkan SPAWN death cases PASS, while the
DirectVulkan inproc arm still fails. So the two-process Magma death path works today and the
remaining gap is the inproc arm alone. Recorded in the package note; narrowing that skip guard
is package C's call and its file.
2026-09-22 14:24:19 -04:00
swung0x48 3d9a38a979 [MG_IntegrationTest, CI] (P7): the split arms take a backend, so Magma has two-process lanes at last
P7 exit gate 1 asks for integration green "under Magma push and split". Until now it was not
merely unmet, it was UNMEASURABLE: every arm of mgl_itest_register_split_arms hardcoded
MOBILEGL_BACKEND_TYPE=DirectGLES (:2002/:2025/:2031) and MGL_ITEST_MAGMA_WIRE_ENVIRONMENT
hardcoded MOBILEGL_TRANSPORT=inproc (:2456), so the shared integration-split / -spawn / -tcp
labels held ZERO DirectVulkan entries and the 73 integration-magma-split cases only ever ran in
one process.

The registration macro now takes the backend as its first argument; the old name is a thin
wrapper, so all 32 call sites are untouched and every DirectGLES name, label and private log
path is byte-for-byte what it was. Two tiers come out of it, and the CMake comment states both:

  * GATED - integration-magma-{split,spawn,tcp}. The six hand-laid Magma lanes (Fm, Buffers,
    RunAhead, Caches, NamedBlit, P5fRsp) grow arms through one shared shape macro; the inproc
    arm keeps its names, its labels and its log paths, so the CI steps and the two magma_*
    checker scripts that name them by hand keep working unchanged.
  * INFORMATIONAL - integration-magma-all-{split,spawn,tcp}, in no gating step. Every scenario
    the split arms run, on Magma, on all three transports, generated by REPLAYING the macro's
    recorded call list - so a scenario added to the split arms tomorrow joins the Magma census
    the same day. Its job is to be able to go red while CI stays green; a case moves up into the
    gated tier with the commit that retires its blocker.

The label spellings are chosen against ctest -L's regex (ID-131 again): the tier word goes in
the MIDDLE, because integration-magma-spawn-all WOULD have been selected by
-L integration-magma-spawn and pulled ~100 expected-red entries into a gated lane. Sub-labels
mirror it: the inproc arm keeps integration-magma-buffers, the others put the arm word first.

MagmaRunAheadScenario and MagmaWireCacheScenario stay on the inproc arm alone, and that is
MEASURED rather than assumed: with all three arms registered, all 21 entries red on spawn and
tcp with no Fatal in either role log - the first arms its hold through
ServerLoopInstance().SetBeforeRetireHookForTesting, a hook on THIS process's server loop, and
the second asserts transportName == "inproc" in its SetUp. Both are named in the CMakeLists and
in spawn_lane_parity.py rather than allowed to widen a tolerance.

split_log_paths.py's is_split prefixes gain DirectVulkan.Spawn. / .Tcp. - on those arms two
processes append to one path, so a shared log does not merely blur attribution, it interleaves
two sessions. spawn_lane_parity.py gains the Magma dimension, normalising only the ARM segment:
RunAhead.Credit1. and .Credit3. are two entries of one case and a tail-stripping normaliser
would fold them together and then report an equal-sized set that had lost one.

Measured on this tree (lavapipe): magma-split 73 unchanged and green, magma-spawn 52/52,
magma-tcp 52/52, the three informational lanes 102 each, GLES split 186 / spawn 102 / tcp 104
all unchanged and green. G14 against ~/w7/p7-before: 4155 to 4565, removed 0, added 410.
2026-09-22 14:24:19 -04:00
swung0x48 9e8fdacd21 [MG_Remote] (P7): gate 3 is judged per case against the monolith arm, and its denominator names what monolith already fails
Device window #1 E0a: monolith x DirectVulkan on the Redmi is 0.996-1.000 against the
desktop goldens (only OpenRA is 1.0), so "monolith 1.0" was an OpenRA fact, not a rule.
Gate 3 now reads: per case, inproc and spawn bit-identical over three repeats, within
the case threshold, and within 0.0005 of the same-session monolith reading. The
denominator excludes what monolith itself fails - create-indirect (OQ-17), the
1.21.11 main-menu trace artifact, and iris-photon, whose pipelines Adreno 830 refuses
with VK_ERROR_UNKNOWN (PipelineFactory.cpp:660) - which leaves 36. Gate 5 records the
real block sources: UBO lives only in the GTF module, SSBO joins, gl33 is the fallback.
2026-09-22 14:23:00 -04:00
swung0x48 8b02feb245 [android-plugin] (P7): the result verdict hands native python a host path whatever the caller set MSYS_NO_PATHCONV to
On Git Bash the script protects its own adb calls with MSYS_NO_PATHCONV=1, but the
two python invocations relied on the caller NOT having exported it; a session that
did (for the adb-side /data/... arguments the runbook mentions) got json.load failing
on /c/... for a result.json that existed - every case reported "trace replay failed"
over a passed result. host_path_for_adb already exists for exactly this; use it.
2026-09-22 14:17:42 -04:00
swung0x48 c890f0ad1a [docs] (P7): the runbook spells the side-by-side package the way the flavour builds it, and its two shell traps are removed
The application id is top.mobilegl.plugin.p7w1.trace (flavour suffix after the id
suffix), pgrep -f dex2oat matched the shell running the loop and never exited, and
adb.exe cannot stat an MSYS /c/ path once MSYS_NO_PATHCONV is set. All three were
hit on the first execution of section 3.
2026-09-22 14:15:29 -04:00
swung0x48 9be62cbc21 [tools] (P7): the UBO caselist comes from the GTF module, and the SSBO block joins gate 5
KHR-GL46 has no uniform_buffer_object group; the 90 UBO cases are
GTF-GL46.gtf31.GL3Tests.uniform_buffer_object*, present only when the deployed glcts
was built with GTF - the device probe decides. KHR-GL46.shader_storage_buffer_object
(124) is the block that exercises what cluster A actually refuses today: unaligned
and over-range SSBO bindings, not only UBO byte tails.
2026-09-22 14:03:58 -04:00
swung0x48 4e7989f1c2 [tools] (P7): the five CTS caselists for exit gate 5 are committed with their provenance
Cut from VK-GL-CTS 0b04c470ed78569b0b778ece1262d88455ab2b00 (main; the deployed glcts build's source) mustpass
gl46-main.txt and gl33-main.txt: texture_*, shader_image*, packed_pixels,
direct_state_access, uniform_buffer*. The gl33 variants are the fallback the runbook
names for an Android runner that cannot open a 4.6 core context; whichever version the
probe selects, BASE and AFTER read the same list. A 0.5 pp gate needs a committed
denominator, not a glob over someone's checkout.
2026-09-22 14:02:07 -04:00
swung0x48 484a56a5ae [scripts, MG_Pipe, MG_Backend, MG_Test] (P7): the census docstring's triage follows what the same wave retired, and the comments stop joining the @P7 census
Two corrections to the two commits before this one.

The widening's docstring was written against the 83 sites it found, and the commit that
routed five of them through Session::Fail did not revisit the prose. It now describes the
79-site floor the baseline actually records, says which five went and where, and states that
the enforceable artefact is the baseline's per-file map rather than the counts in the prose -
so a reader who finds the two disagreeing knows which one to believe. The three buckets are
restated too: `(ii)` is no longer "owned by a Ph / P7 row", because with the Magma wire arms
routed what is left there is the peer-reachable deaths in the applier and the two backends,
which is exactly the set ID-P7-1 scopes to Ph's fuzz arm 2 AFTER P7.

And the routing commit's own comments spelled the at-sign refusal marker while explaining
what routes through them, which put SEVEN comment lines into the retirement census: `grep
@P7` went from 15 sites in 4 files to 22 in 6, and that grep is how wave 2 will measure which
refusals it has retired. The prose says the word instead, and the count is 15/4 again.
2026-09-22 14:01:06 -04:00
swung0x48 a5dcbb444f [MG_Remote, scripts, MG_IntegrationTest] (P7): the supervisor names a faulted session, and the smoke that proves it is finally a ctest entry
Ph slice (1) (ID-P7-1). Three `::waitpid(active, nullptr, WNOHANG)` calls in the `--serve`
loop threw the status away, so a session child that died of SIGABRT inside SessionFail and
one that returned from its control loop and _exit(0)'d were the SAME EVENT to the
supervisor: `active = -1`, no line, nothing to count. ID-P7-1 declines the literal 98-site
latch translation because P6.5's fork-per-session already delivers "the next connection is
served as usual"; what was missing is that anyone could TELL a faulted session from a
finished one, which is what an operator, a fuzz arm and a flake triage all actually need.

One `ReapActiveSession()` where there were three call sites, asking WIFEXITED / WIFSIGNALED /
WTERMSIG - the same three questions ServerSpawn.cpp:289 asks of the same kind of child - and
logging `pid=<n> reaped exit=<c>` or `signal=<n>` with the supervisor's lifetime
`sessionsFaulted` beside it. A refused handshake exits 0 and is NOT a fault, which is right:
protocol.fbs:86 says a rejection is a normal session outcome. The running total is on every
reap line as well as on the summary, because the fixtures end a supervisor with SIGTERM and a
figure only printed at a chosen exit would never be seen.

THE IDENTIFIER ROW'S PREMISE IS WRONG, MEASURED. The plan asks for
CtrlEnvelopeBufferHasIdentifier at ServerMain.cpp:193 because ServerSession::ParseEnvelope
asks it and this loop did not - but the generated VerifyCtrlEnvelopeBuffer already IS
`VerifyBuffer<CtrlEnvelope>(CtrlEnvelopeIdentifier())` (protocol_generated.h:2378), so a
wrong identifier was refused here and ParseEnvelope's second call is redundant too. Recorded
for the integrator. What was real is that both refusals were SILENT: a peer whose
post-Welcome frame was garbage had its session ended with no line in the log, and a fuzz arm
that cannot tell "refused, by name" from "the child went away" measures nothing. So the
identifier is now asked first - it is the specific diagnosis where the verifier's is the
general one - with the length guard that asking it first requires, and both say so.

scripts/ci/tcp_supervisor_smoke.py has been in the tree since P6.5 asserting four supervisor
controls and was referenced by NOTHING - no ctest entry, no workflow step - so nobody could
have said whether those four were green. It is now `TcpLane.SupervisorProtocolControls`
beside the TCP lane's own fixtures, labelled integration-tcp, under the lane's resource lock
(a second supervisor's child would answer the other's peers with Busy). It needs flatc, which
is deliberately not in the build graph, so it resolves one gen_protocol.py's way and exits 77
- ctest SKIP - when there is none, rather than passing on a run that did not happen.

Two cases added to it: a session child killed with SIGABRT must be named by pid and
`signal=6` with `sessionsFaulted` one higher than before, and the next Hello must still be
welcomed; and a steady-state frame carrying the wrong file identifier must be named rather
than silently dropped.

Red-once, both ways, against the real image: making the reap discard the status again reds
with "the supervisor did not name session pid=N as killed by signal=6", and deleting the
identifier branch reds with "read a frame with the wrong file identifier without saying so".
2026-09-22 14:01:06 -04:00
swung0x48 e1ec69341a [MG_Pipe, MG_Backend, MG_Remote, MG_Test, scripts] (P7): the Magma wire fatals and the two Staged aborts go through Session::Fail
Five deaths in the split server image raised their own std::abort(). Three of them are
funnels - MagmaWireFatal (Renderer/WireFramebuffer.inc), WireDescriptorFatal
(Renderer/UniformManager.cpp) and WireBufferLegacyFatal (Renderer/WireDraw.inc) - and the
fifteen `@P7` refusals on the tree (8 in UniformManager.cpp, 4 in WireFramebuffer.inc, 2 in
WireDraw.inc, 1 in VulkanRenderer.cpp; plan section 1.3's count confirmed) all die through
two of them. Every one of those deaths published NO SessionFault frame, so the peer read a
bare EOF; bumped NO SessionFaultCount(), so exit gate S8 could neither confirm nor falsify
it; and carried a hand-typed family word with no path to FatalFamilies.def's projection.

MG_Backend may not name an MG_Remote symbol on its own account, so the route is a seam in
MG_Pipe with MGP_TRIP_WIRE_REPORT's shape (PipeApply.cpp:66, and the new code sits beside
it): MG_Pipe/PipeSessionFail.h declares MGPipeSessionFail, PipeApply.cpp defines it inside
`#if MOBILEGL_BUILD_DISAGGREGATED`, the DEFAULT behaviour is byte-identical to what the
three funnels did on their own, and MG_Remote::InstallPipeSessionFailHook() points it at
SessionFail. The install is in InitServerRoleCommon (MG_Backend/Init.cpp), step 0, because
that is the one function BOTH server roles run - the inproc role and the spawn/TCP session
child - so there is no per-transport install to get wrong.

MGPipeFatalFamily is a two-value boundary enum, NOT MGFatalFamily cast to an integer: the
.def's enumerators are positional and a row inserted in its middle would silently re-point
every hook call at a different family. FatalFunnel.cpp's adapter is the one place the two
vocabularies meet, as a switch with a named default.

The family word stays UnmigratedVerb rather than becoming a new MagmaWireRefusal. Three
reasons: the message strings are required to stay byte-identical (5.2), and the retrace
refusal censuses recorded since P5b count these deaths by that word - a rename invalidates
every `@P7` measurement taken against them; `Magma:` in the detail already separates the
Magma arm from the GLES UnmigratedVerb deaths, so a new word adds no discrimination a grep
does not have; and its wire projection would be ProtocolCorruption, exactly UnmigratedVerb's.
Recorded for the integrator.

Ph slice (2) in the same commit: Server/StagedShadow.h's RequireCovered and
Server/StagedTextureStore.h's RequireLevelBytes were MGLOG_F + bare abort. Both are inside
MG_Remote already, so they call SessionFail directly, message verbatim;
check_include_closure.py's four probes do not reach either header and the gate is green.

Census: 83 abort sites -> 79 over 20 files (five sites retired, one gained for the seam's
no-hook default, which is argued into FUNNEL_SITES), baseline rewritten in this commit - the
downward ratchet's notice is what asked for it. Two gate bugs the widening exposed and this
commit fixes: the SessionFail call regex also matched `MGPipeSessionFail(` (a lookbehind
now), and a funnel's anchor needed a wider window than a marker's (kFunnelAnchorWindow).

Red-once, three ways. Pointing MagmaWireFatal back at its own abort reds the census naming
WireFramebuffer.inc 0 -> 1. Making the seam skip the hook reds both new death cases -
`TheMagmaWireVerbDeathReachesSessionFailAndBumpsTheFaultCount` on the exit code, because it
reads SessionFaultCount() out of a SIGABRT handler and gets 0 instead of 1, and
`TheSeamHandsTheInstalledHookItsFamilyAndItsLineVerbatim` on the line the hook echoed.
2026-09-22 14:01:06 -04:00
swung0x48 f141918aa1 [scripts, MG_Remote, MG_Impl, MG_Test] (P7): the fatal census gate widened to the split server image
`scripts/ci/fatal_census.py` scanned `MG_Remote/` and saw four abort sites. CONTRACT-P6
§12.1 already says what the server image is - "P6's server image is the whole
libMobileGL.so" - and the server process was executing seventy-nine more deaths the gate
could not see: 15 in DirectGLES.cpp, 12 in Managers.cpp, 11 in VkBufferManager.cpp, 7 in
PipeRoute.cpp. SCAN_ROOTS is now MG_Remote / MG_Pipe / MG_Backend / MG_Impl/Pipe /
MG_State, `.inc` joins `.cpp`/`.h` (the Magma wire arms and the generated wire records are
.inc files, and 20 sites are reachable only through one), and MG_Benchmark /
MG_IntegrationTest join MG_Test in the exclusion.

Four rules now, where there were two:

  1. an abort names its family within twelve lines, OR it is argued into FUNNEL_FILES /
     the new FUNNEL_SITES table. FUNNEL_SITES exempts an individual funnel inside a file
     that has ordinary named deaths around it, and only when one of the file's anchors
     appears in the same window - so the exemption follows the funnel if it moves, expires
     if it is deleted, and does not cover an unrelated abort added to the same file. Two
     entries, one line of reason each: PipeApply.cpp's MGP_TRIP_WIRE_REPORT and
     PipeWire.inc's MGPipeWireProtocolFatal.
  2. unchanged: a WireLogFatal / SessionFail call carries a `Fatal{` in its string.
  3. NEW: every family word found in the perimeter has a row in FatalFamilies.def. Eleven
     words did not, and now do (see below).
  4. NEW: every `Refuse{Word}` names a RefuseCode enumerator. This is what catches
     `Refuse{AuthenticationRequired}` (SocketTransport.cpp:161), a word protocol.fbs has
     never had - PH-7 (2). Renamed to the enum's own `Authentication`, and
     SocketTransportTest pins the spelling to `EnumNameRefuseCode` so the two cannot
     drift again.

`abort_sites` was recorded and never compared. It is now a DOWNWARD ratchet with a
per-file map beside it, so "83 became 84" reports as "VkBufferManager.cpp went 11 -> 12,
here are its lines"; fewer sites - the direction `@P7` retirement moves the number - asks
for a re-baseline rather than blocking one.

Eleven new `.def` rows for the eleven rowless words. Ten are ProtocolCorruption for the
reason that bucket already states. ResourceUnavailable is ServerCrashed, not SegmentMismatch:
SegmentMismatch means a shared SEGMENT was the wrong size, and telling a peer that when a
VkBuffer allocation failed would be a lie about whose geometry was wrong. A twelfth word was
collapsed rather than given a row - `InitialBytesNotCarried` (PipeFill.cpp) said the same
thing as `UncarriedInitialBytes` (WireTables.cpp) about the same record, and two words for
one family is the drift a6 censused.

Red-once, four ways: an unnamed `std::abort()` appended under MG_State reds naming the file
and the line; deleting the `Fatal{` from PipeRoute.cpp's ResidualBlockSize reds; putting
`Refuse{AuthenticationRequired}` back reds rule 4; a typo'd family word reds rule 3 and the
vocabulary ratchet together.
2026-09-22 14:01:06 -04:00
swung0x48 1e2d076633 [docs] (P7): the wave-1 device window runbook
One reboot-clean session on Redmi 2f7cbe2e, copy-pasteable, with the shell each
block runs in marked: pin and wake/Doze, APK install identity, P7-7 E0-E3, the
P6.5 residual 39-case DirectGLES TCP golden matrix, the CTS $BASE for monolith x
DirectVulkan, and the temp-environment restoration. Every step states in one line
which result supports which hypothesis, and where the evidence is archived.

Four things the runbook had to settle rather than restate:

E1's proof. `running lockstep` is NOT the line that proves lockstep took effect:
ClientSession.cpp:1873 guards it with `RunAhead != 0 && m_barrierArmed`, so it is
the "the knob is set but the server lacks kCapRunAheadApply" warning and
MOBILEGL_IPC_RUN_AHEAD=0 prints nothing at all from LatchRunAheadFromCaps. The
proof is a line present and a line absent: `Config: IPC ... run-ahead=0 ...`
(ConfigLoader.cpp:459) with zero `run-ahead ARMED` - the shape p5f/magma-runahead.md
already recorded for the RA=0 arm.

E1 will exit non-zero, by design. --transport inproc adds --require-inproc, whose
gate demands run-ahead=1 in the Config: IPC line (trace-replay-ci.sh:601), and the
proof runs AFTER result.json is pulled and checked. So the SSIM and the archive are
usable and the arm is verified by hand against the table above; dropping the --env
to make it green would delete the experiment.

E0's budget. 39 cases x 2 transports x 3 repeats is 234 replays, roughly 3-4 h,
which does not fit beside the TCP matrix and CTS in one window. It is split into a
one-pass census that produces the divergent-case list and a three-repeat pass over
that list plus two controls - the determinism question only has meaning for cases
that already diverge.

CTS is flagged, not promised. No caselist is in-tree, and the local VK-GL-CTS
checkout has an empty external/openglcts/data with no mustpass file anywhere, so
the five blocks cannot be cut in-tree; the exact pre-window procedure to fetch a
tagged checkout and commit tools/cts/caselists/p7-*.txt is written out instead.
And KHR-GL46 is unproven on the Android runner - every device reference result in
the skill is KHR-GL33, the 4.5 numbers are desktop lavapipe, and the platform hands
the package's version straight to eglCreateContext - so the runbook carries a
five-minute pre-window probe and an explicit integrator decision: if 4.6 is
unreachable the baseline is taken on KHR-GL33 and the AFTER reading must use the
same version, because 0.5 pp between two different API versions means nothing.
2026-09-22 13:58:07 -04:00
swung0x48 3ba150d021 [tools, android-plugin, CI] (P7): the device matrix runner selects DirectVulkan, repeats are archived, and the spawn env parse is entry-aware
wave 0 item 5. Three instruments P7's device window needs, and the CI steps that
can make each of them red.

run_tcp_matrix.py --backend DirectVulkan end to end. --backend already existed and
already filtered by each case's ci_backends; what did not exist is a per-backend
GOLDEN. The two backends shared `golden` plus an optional `alternate_golden`, and
alternate_golden is an OR across backends, not a per-backend key - so a
DirectVulkan run that matched the DirectGLES picture passed, which is the "went
green on the wrong arm" shape the exit gate exists to catch. trace_cases.json
gains `backend_overrides.<backend>` ({golden, alternate_golden, ssim_threshold}
only: restating target_call or a crop would stop the two backends retracing the
same frame through the same window). A per-backend golden REPLACES the shared
pair, matching another backend's declared image is refused BY NAME, and a --case
named explicitly that does not run a requested backend is an error instead of one
plan fewer and no message. The CMake emitter and the APK matrix refuse a case that
declares the key rather than silently emitting one golden for both arms.

DirectGLES is unchanged and that is asserted, not asserted-in-prose: a case with no
override resolves to itself, and the key is dropped from the resolved case even for
the backend that overrides nothing - so giving DirectVulkan its own golden does not
move the DirectGLES identity fingerprint and does not invalidate its checkpoints.
Each results.json row now records the golden, alternate golden and threshold it was
scored against.

run_android_retrace_local.py gets --matrix (the CI split subset, 39 cases, each
running only its own ci_backends - not --all, which adds the rd12 workload ID-P7-4
excludes), --repeat N, and --archive-dir. The archive is the point: the result root
is keyed by case and backend and by NOTHING else, so repeat 2 overwrites repeat 1
and the only artefact that survives a three-repeat run is the last one, which makes
"were the three repeats the same picture?" unanswerable after the fact - in the one
place where re-running to find out is expensive. It also refuses --benchmark with a
non-monolith --transport, which has always produced a monolith run under a spawn
label (run_benchmark_case passes args.env alone; p6/README.md records the pitfall).

compare_actuals.py computes the pair result.json cannot: SSIM between two actual
PNGs, plus a case x repeat x ssim-vs-golden x ssim-vs-first-repeat table over an
archive. It is the retrace gate's OWN SSIM - a transcription of ComputeChannelSsim
/ ComputeRgbSsim (trace_replay_core.cpp:630/:671), the only SSIM in the tree - and
the crop comes from each repeat's own result.json, so the two columns describe the
same rectangle. PNG decode is stdlib zlib, because an unavailable Pillow in the
device window would turn "the repeats diverged" into "the tool did not run".

TraceReplayActivity: resolveSpawnServerPath's two decisions were String.contains()
on the whole joined K=V;K=V list and neither was entry-aware. The one that bit:
`--env MOBILEGL_IPC_SERVER_PATH` with no '=' is the documented UNSET spelling
(trace_env_overrides_test.cpp:92) and the only way to ask for "spawn with no image"
and watch LaunchServer refuse by name - it did not match "MOBILEGL_IPC_SERVER_PATH="
so a resolved path was appended AFTER it and, because ApplyEnvOverrides applies in
order with setenv(...,1), the deliberate unset was silently undone. That is exactly
the "silently replacing it would make that knob untestable" the method's own comment
forbids. Two more: the key inside another entry's VALUE flipped both decisions the
wrong way, and the transport match was case-sensitive while ConfigLoader lower-cases
the value before matching (ConfigLoader.cpp:322). The parse moves to
SpawnServerPath, a class with no android.* import, so test_android_lifecycle.py
compiles and runs it with plain javac - the parse had no host-side coverage at all,
and its on-device failure mode is "the knob had no effect".

Only spawn gets a path, deliberately: it is the only transport that execs an image.

Red-once, all four: a stub case whose result names the DirectGLES golden fails a
--backend DirectVulkan run with "matched a golden another backend declares"; a
synthetic PNG pair with a closed-form SSIM (identical = 1.0 exactly, constant
black vs white = C1/(255^2+C1), inverted halves = (C2-2v)/(2v+C2)); the pre-fix
resolveSpawnServerPath body fails eight of SpawnServerPathTest's checks; and the
DirectVulkan TCP proof battery is the DirectGLES one under the other prefix,
because an empty prefix selection raises nothing and reads as a clean pass.

CI: test_run_tcp_matrix.py and test_compare_actuals.py join the R-16 step. The
matrix driver's stub regressions had NO CI step - they were run by hand and
recorded in a note - and they guard the tool that decides whether a device golden
run passed.

Counts in WSL: test_run_tcp_matrix 12 -> 18, tcp_lane_tools 4 -> 6,
test_compare_actuals 12 new, test_android_lifecycle green with the new Java unit.
2026-09-22 13:58:07 -04:00
swung0x48 e19095269b [docs, MG_Remote] (P7): the roadmap rows say what the tree says, and CONTRACT-P7 is drafted
ROADMAP.md: the Ph / P3b-P4b / P7 rows carry the audit's verdicts and the factual
corrections it found - the @P7 refusal list is 15 sites in 4 files (the row printed 8),
the 29-commit DirectVulkan shape wave was never recorded, the device divergence is two
observations with no per-case attribution and no lockstep control, SetupDrawSnapshot /
content-addressed CSO are declined for this stage, "XFB scatter to client" is retired
in favour of the two real holes, readback/pack state splits into closed and open
halves, the NDK size pin is obsolete, PACK-PBO readback landed in 70df57bb, the ABI
fingerprint landed in P6.5, and OQ-8 / OQ-10 / OQ-13 / OQ-21 get their current answers.
P6-ENDSTATE-REVIEW.md:284 qualifies an Init.cpp citation the lint reported ambiguous.

MG_Remote/CONTRACT-P7.md: rules I (no std::abort on the Magma wire arm) and J (every
fix reds once on a two-process arm), the five exit gates in testable form, the
retire-vs-decline table for every @P7 name, the ratchet's definition and P7 target,
the nine-subsystem denominator for the 40% checkpoint, the wave-2 file partition,
and the amendments to CONTRACT-P6. Census, ratchet and lane numbers are marked
待填 until wave 0's packages report them.
2026-09-22 13:26:34 -04:00
swung0x48 e8b2c4bd05 [docs, MG_Test] (P7): the Ph / P3b-P4b / P7 program plan, and the pull flavour builds its tests again
Plan for driving the roadmap to P7 close-out, from a per-item audit of the three
rows against a96875bf (19 reader groups, two refuters per done/obsolete verdict,
one completeness critic): what is landed, what is declined and why, what remains
with file anchors, five waves, the P7 subsystem denominator for the 40%
re-baseline checkpoint, and eleven integrator decisions (ID-P7-1..11).

The three findings the roadmap text did not carry: Magma has zero two-process
integration coverage (every split arm hardcodes DirectGLES), the 184-symbol
ratchet has no instrument, and the device DirectVulkan divergence was never
diagnosed (the lockstep control was never run under golden).

MG_Test/Util/PipeStatsTest.cpp: the pull flavour (MOBILEGL_PIPE_PUSH=OFF) did not
compile since the P6 wrap-up - the window-arithmetic case used the push-only
PipeStats::Gauge API unguarded. CI's pull control builds with
MOBILEGL_BUILD_TEST=OFF and never saw it. The gauge lines are now #if
MOBILEGL_PIPE_PUSH; the case name stays registered in both flavours for G2/G14.
2026-09-22 13:03:20 -04:00
swung0x48 b95f5f1a20 [docs] (P6.5): the first wave recorded as landed - review + fixes + representative re-validation, residual to CI 2026-09-22 11:58:22 -04:00
swung0x48 fe28bdb5f3 [MG_Remote, MG_Pipe, MG_Test, CI, tools, docs] (P6.5): the first wave completed and reviewed - archive v2, publication boundary, half-close, AcceptPair/role fixes, device re-validation 2026-09-22 10:56:19 -04:00
swung0x48 018fab0f5b [WIP] (MG_Remote): implement P6.5 TCP and stream transport 2026-09-22 07:19:18 -04:00
swung0x48 242fd45cd7 [Fix] (CI): read role logs and retry device disconnects 2026-09-22 07:19:17 -04:00
swung0x48 9043442454 [MG_Test, CI] (P6): the last two readers the role split left behind
build-linux-verify could not compile PipeStatsTest: RoleDerivedJsonDumpPathForTesting is DISAGGREGATED-only and the verify flavour is deliberately monolith. The case name stays registered and GTEST_SKIPs there, matching the PIPE_PUSH pattern, so ctest -N still matches between flavours.

runtime_mode_proof.py logs opened MOBILEGL_LOG_FILE_PATH as a file. After the split that name is a base and the sink writes <base>.client.log / <base>.server.log, so the inproc proof died on FileNotFoundError after a green lane. It now reads through split_log_paths' role helpers and names the paths it looked for when none exist.
2026-09-22 04:40:03 -04:00
swung0x48 35b3283bd9 [docs] (P6): the closeout - gates 7 and 8 are measured and P6 is closed
The device gate: Redmi 2f7cbe2e, reboot-clean, paired A/B in one thermal
window (arms interleaved, best-of-3 then best-of-2), rd12 + DirectGLES.
inproc vs spawn client CPU p50 7.880/7.889 vs 7.876/7.869 (-0.15%) and
7.987/7.989 vs 7.992/8.009 (+0.16%) - a tie inside the 0.2-4.2% within-
arm spread, sign flipping with arm order, recorded and not gated.
monolith is ~30% dearer. 22/22 runner exits clean, 44/44 PINNED, zero
Fatal.

The three mandatory numbers: socket doorbell vs condvar is a tie too -
client 121.1 waits/frame identical across transports, server 10 866-11
963 (inproc) vs 10 415-10 416 (spawn) waits/frame, and the contract's
'converting that into ~4 blocking reads may be cheaper' did not pay off;
the only column where the socket is worse is clipark/f (6.07 vs 5.60/5.75).
SEG_STAGE bytes/frame: 130.8 KB median / 8.01 MB steady (OpenRA), 816.2
KB / 1.53 MB (rd12). Records/frame post-chunking: 211/206 and 7 497/7
504. maxrec stays 1808 - the 8 MiB chunk budget is never reached. Both
framings reported throughout (the startup window runs ~7x the steady
state) rather than the flattering one.

MEASUREMENTS.md gains section 12 with the full protocol and tables; the
6.3 'no post-chunking distribution exists' gap now points there.
CURRENT_STAGE_PROGRESS.md marks gates 7/8 met and records the closeout
work plus two structural limits (spawn client windowed fields are zero
by construction - OnPresent runs in the server process - though its run
totals are valid stores; the stats dump is role-derived). ROADMAP marks
P6 closed in the stage table, the header and the milestone.

Evidence: notes/p6/gate7-device-ab.md, gate8-doorbell-device.md,
gate8-wire-counters.md, git-worktree-fix.md. check_doc_citations: 82
citations in the three documents, 0 problems.
2026-09-22 03:38:17 -04:00
swung0x48 06b61bd84e [MG_Util, MG_Remote, tools] (P6): the wire counters and the spawn server's stats latch
Gate 8's last two numbers (SEG_STAGE bytes/frame, records/frame post-
chunking) had no producer at all: the summary line's byte classes are
backend-to-driver bytes and maxrec is a max, not a per-frame. Two push-
guarded counters answer them at their single choke points - ByteClass::
StageSegmentBytes at PipeWireCodec's StageAllocate (the only allocation
behind StageBytes, so every buffer walk, texture slab and CSO archive
lands there and emitters cannot double-count) and CallClass::WireRecords
on EncodeRecord's committed path (post-chunking an emitter no longer
knows how many records its one call produced). A staged-blob histogram
rides along. G1 holds: symbols 0/0/0/0, .text byte-identical, pull build
carries no MG_Remote symbols.

The spawn server never counted anything: PipeStats::Init() is a step of
MobileGL::Initialize(), which ServerMain never reaches, so every Enabled()
site in that process - including ServerLoop's own ServerWaits/ServerParks
publish - stayed false and the server log carried no summary line at all
(a zero that means 'another process' printed in the shape of one that
means 'never waited' is the milder misreading; the line was simply
absent). ServerMain now arms and closes PipeStats at its role boundary -
after MG_ConfigLoader::Init (Init latches from the parsed config), after
the server-role latch (the log sink's role), and before ::_exit(0).
Red-once: 0 server summary lines before, 29 after, SSIM 1.0 unchanged.

MOBILEGL_PIPE_STATS_FILE now derives per role through the log sink's
exported RoleLogPath rule - run.client.json / run.server.json, the bare
name resolves to no file - because both roles reach FormatJson() and ofstream
trunc made the second writer replace the first. G1 caught three separate
leaks on the way (a string temporary in the monolith body, one resized
Shutdown() from unequal #if arms, two rewritten MGLOG_W literals in
.rodata); the pull arm is statement-for-statement the original.

Device tooling for the paired A/B lands with it: pin_device.sh gains a
measured 2f7cbe2e entry (this board clamps the GPU at 1050 MHz -
thermal_pwrlevel writes 0 and reads back 1, max_gpuclk is read-only;
absolute times are not comparable with the 1100 MHz-era rows), and
tools/device_bench/p6/ carries the session collector and renderer beside
run_android_retrace_local.py's MOBILEGL_TRACE_PACKAGE override.

Unit 2367/2367 (PipeStats 25/25), spawn integration 102/102.
2026-09-22 03:37:51 -04:00
swung0x48 4e7bf0fe65 [docs] (P6.5): the design for the first wave - all-TCP transport, WSL client to adb-device server
ROADMAP gains the section "the immediate next item": a design, not an
implementation, for running the frontend on an x86_64 WSL process and the
backend on the Redmi over TCP, with the integration lane and the CI retrace
subset as the acceptance. The P6.5 row now points at it.

What the design rests on, verified in the tree: the control plane already
treats the rendezvous as a NAME with two ordered connections (SocketTransport),
ClientSession::StartOverSocket already separates "how the socket came" from the
handshake, MGPBlobRef addresses kSegStage only through one ResolveOrFatal, the
reply header and the event heads are already wire-shaped, LogLine exists in
CtrlMsg with no producer, goldens are shared between desktop and device, and
the trace APK can already exec libMobileGLServer.so from nativeLibraryDir.

What it does not rest on, stated rather than assumed: ILink is declared and
unused beyond LinkProgress, ShmLink does not exist, both sessions hold
SessionSegments directly - the lk seam body from the end-state review was never
built and is this wave's first package (lk2). The manifest has no INTERNET
permission.

The mechanism: StreamLink as mirrored private segments replicated by messages -
same sizes both ends, same offsets, so the codec, the watermark rules, the apply
thread and the encoder do not change; RingControl becomes Progress frames sent
on flush-on-idle (the recorded amendment to "appliedSeq is never lazy": late by
at most one flush, never early); replies become seq-stamped messages; the event
ring is mirrored with the P5e blocking rule intact. Cross-architecture makes
the PipeFields.def layout digest mandatory (wf). Server logs on another machine
make LogLine forwarding mandatory (lf) so the 34 server-log readers stay
unchanged. A --serve supervisor forks one child per session on the device,
started by a foreground Service in the trace APK.

Packages lk2 -> wf -> ct -> sl -> lf -> sv -> t65, each with red-once; the gate
list, the mandatory numbers (kWaitReply per frame x RTT, SEG_STAGE bytes/frame
against the link, PRESENT_CREDIT 1..3), the two-sided configuration runbook and
the risks (RTT per reply row, single large uploads, AP isolation) are in the
section. check_doc_citations: 74 citations, 0 problems.
2026-09-22 03:04:50 -04:00
swung0x48 872370ba9d [docs, MG_Remote] (P6->P6.5): the end state is TCP across machines, and the transport stack has two axes
The end state is re-ruled: client and server may sit on different machines,
OSes and architectures, joined over TCP. The AVF pVM / AF_VSOCK end state the
P6 end-state review assumed is withdrawn (retraction at its head, form D added
to its table, the vsock probe and both external unknowns struck).

The transport stack becomes two independently selectable axes, negotiated in
the handshake and mixable: control plane (ITransport: fork / unix:<path> /
tcp://host:port) and data plane (ILink: ShmLink / StreamLink). On one machine
control over TCP with data over shared segments is a legal pair; ShmLink brings
its own AF_UNIX aux rendezvous for SCM_RIGHTS, so ITransport narrows to control
frames. `auto` is decided by a successful segment delivery, never by inspecting
addresses, and a fall-back is named. Nothing above Transport/ may branch on the
link kind. ARCHITECTURE.md gains 11.9 for this; CONTRACT-P6.md gains 10.3.

ROADMAP: P6.5 and Ph get rows (both were referenced by the contract and absent
from the stage table); the IPC track becomes P6 -> P6.5 -> Ph -> P12 -> P9 ->
P10 -> P11(same-machine arm). Every later row is reconciled against the tree:
items already landed are struck (client vertex arrays, client-minted query
handles, credit default 1, SEG_EVENT overflow mechanism), the unreachable P7
gate (`nm | grep glslang` on a server image that IS libMobileGL.so) is replaced
by the 184-symbol ratchet share, fixed-width wire form moves from P7 to P6.5,
P12 is cut down per the review (server-owned display, no Surface passing), and
the debt table's stale claims are corrected: class C is down to two slots, not
15 and not 511; PACK-PBO and the 2 MiB reply cap are P9's, not P6's.

Verified against the tree and recorded honestly: the contract's three hs rows
(LinkTerms, Refuse, the fingerprint split) never landed - protocol.fbs has
neither - so CURRENT_STAGE_PROGRESS.md's claim that S6 ran red is withdrawn and
the rows move to P6.5; the 184-symbol ratchet gate does not exist in CI; the
server process never initialises PipeStats, so gate 8's `srv=0 srvpark=0` is a
structural zero (fix in progress in the tree, not in this commit).

check_doc_citations.py: 159 citations, one pre-existing basename ambiguity.
2026-09-22 02:17:18 -04:00
swung0x48 94c130d639 [MG_Util, MG_IntegrationTest, CI] (P6): the readers the role split left behind
The per-role log split renamed BOTH halves so that an un-updated reader would
fail loudly instead of silently reading half a session. It worked: CI found
four readers this way. This is them.

- MG_Util/Debug/Log.h: the role vocabulary now exists in the PULL build too,
  header-only. It was wholly inside #if MOBILEGL_BUILD_DISAGGREGATED, so
  PipeInputsTest - which builds in the verify flavour - could not name
  ReadRoleLogs and took the whole build-linux-verify job down at compile time.
  Inline rather than out-of-line because of G1: nothing in the library calls
  these, so an inline definition emits code into the readers and NOTHING into
  libMobileGL.so. Verified: pull symbols 30570 -> 30570, 0 added / 0 removed,
  .text still 0xa52203, and zero role-log symbols in the pull library.

- PipeStatsWindow.h: the harness copy of the naming rule is now flavour-aware.
  It derived `.client` in BOTH flavours, so in the verify build - one role,
  MOBILEGL_LOG_FILE_PATH written unchanged - it opened a file that never
  existed. The copy has to stay a copy: MG_Util's RoleLogPath is `t` (local)
  in the shipping .so this module links, so calling it does not link.
  Added LogMark/MarkLaneLog/ReadLaneLogSince: one offset PER ROLE, because a
  single scalar over the concatenation slides by whatever the other role wrote.

- The four arming scenarios: they asserted on lines the BACKEND emits, and
  under inproc the backend runs on the apply thread - the server role - so the
  line lands in the server's log. Reading only the client's half reported the
  emulation unarmed, i.e. accused the product of a defect the reader invented.
  Four copies of the same private helper pair now delegate to the harness.

- scripts/ci/magma_runahead_checks.py: it unlinked and read the base name as
  if it were a file, so the negative arm saw an empty log and could not prove
  the knob took. It now imports split_log_paths' derivation - the authoritative
  one - rather than becoming a third copy.

Local verification (WSL, lk-split + c6-pull):
  - the 4 CI-red entries: 10/10 PASS under inproc, 10/10 under monolith
  - unit role-log readers: 510/510 PASS
  - magma_runahead_checks negative: run_ahead_0 EXPECTED_RED, restored GREEN
  - the 5 affected TUs compile clean with MOBILEGL_BUILD_DISAGGREGATED=0
  - G1: 0/0/0/0, .text 0xa52203

Red-once is the CI run itself (35677991453), which failed for each of these
reasons before the change.
2026-09-22 00:37:39 -04:00
swung0x48 75dd0cb1ca [docs] (P6): the roadmap and the current-stage doc record P6 as landed-not-closed
All ten P6 packages (a6/c6/lk/so/sm/cp/hs/dl/st/t6) are on feat/disaggregated
through 16bfab10. The two docs now say so honestly:

- ROADMAP.md (cumulative): the status line, the P6 stage row, and a new
  milestone bullet record every package landed, exit-gate items 1-6 met, and
  items 7 (device paired A/B in one thermal window) and 8 (the three mandatory
  perf numbers) still pending -- so P6 is "implementation complete, exit
  perf/device numbers pending", NOT closed.
- CURRENT_STAGE_PROGRESS.md: rewritten to hold ONLY the current stage. All
  P5* history is gone (it lives in ROADMAP.md and the P5* reports); what
  remains is P6's package table, the CONTRACT-P6 section-9 gate with each
  item's honest status, the device results, the per-role log split, and the
  explicitly-not-P6 list.
2026-09-21 22:02:15 -04:00
swung0x48 77cbd17636 [MG_Remote] (P6/dl): record the Session::Fail funnel as landed in the contract
§5.2 gains a LANDED note: the three steps (the FatalFamilies.def vocabulary,
the SessionFail funnel that byte-identically routes ~90 aborts through one
place, and the SessionFault frame that names the family to the peer), and the
one deliberate deviation from the prose - seq/op are not threaded through the
90 sites because the message already names the op, so the frame carries code,
family and message rather than a field that is always zero.
2026-09-21 21:52:18 -04:00
swung0x48 b46da7002c [MG_Remote] (P6/dl): a dying session names itself to the peer
Step three of CONTRACT-P6 5.2, the funnel's whole point: instead of the peer
reading a bare EOF and guessing, the session publishes a SessionFault frame
that names the family before it aborts.

THE WIRE CHANGE IS D2's, exactly: `table Fatal` appends `family: string`,
append-only, so an older peer still reads `code` and `message`. `code` stays
the coarse seven-value projection; `family` carries the full word.

SessionFail publishes it before the abort, on whichever session's control
plane is active - the client has a ClientSession, the server a ServerSession,
so under spawn the frame reaches the far side whichever role is dying. Under
inproc both exist and it goes to an in-process transport the aborting process
will never read, which is harmless and cheaper than a special case that would
then be the untested one. Best-effort by nature: this thread is about to die,
so a send that blocks or fails changes nothing.

seq/op from the prose are NOT threaded through the 90 sites: the message
already names the op where it matters, so the frame carries what it can name
honestly (code, family, message) rather than a field that is always zero.

The client recognizes it: RunRemoteSurfaceControlFrame's reply loop, which
already demuxes CapsSnapshot from SurfaceReply, gains a Fatal arm that logs the
family and detail, latches device-lost from it, and returns transport-closed.
A server that DETECTS a fatal condition and aborts itself now tells the client
why; a kill -9 still bypasses abort entirely and is caught by dl's hangup
witness instead - the two cover the two ways a server can die.

The census learns SessionFail: a SessionFail call, like a WireLogFatal call,
must carry a Fatal{ word in its string, so the enum and the string cannot
disagree. FatalFunnel.{h,cpp} join the funnel exemption. Baseline is 4 aborts.

FatalFamilyTest gains the wire round-trip: every family's frame carries its
word and its projected code across encode/verify/decode.

G1 re-measured after the schema change: 0 symbol diff, .text unchanged at
0xa52203. Unit 2359 (one pre-existing environmental failure), integration-split
186/186, integration-spawn 102/102, retrace 6/6, census 0 unmarked.
2026-09-21 21:51:50 -04:00
swung0x48 4bb4d4fa4f [MG_Remote] (P6/dl): every session death goes through one funnel
Step two of CONTRACT-P6 5.2. The ~90 scattered `MGLOG_F("...Fatal{...}"); std::abort();`
pairs under MG_Remote/ become one call each to SessionFail(MGFatalFamily, fmt, ...),
which logs, echoes to stderr, counts, and aborts in one place. a6's census
found no single chokepoint existed; now there is one, and step three hangs the
SessionFault-to-peer publish off it.

THE MESSAGE STRING IS PASSED VERBATIM. The contract requires the log lines, the
CI greps and every red-once byte-identical, so SessionFail does not compose the
line from the family - it takes the SAME string the site handed MGLOG_F and
forwards it unchanged. The family enum rides alongside, redundantly, so the
funnel has the coarse wire FatalCode without parsing the string. The census now
checks that redundancy the way it checks WireLogFatal: every SessionFail call's
string must carry a Fatal{ word, so the enum and the string cannot silently
disagree.

The conversion was mechanical - 89 sites by a transform that injects the family
as MGLOG_F's new first argument and drops the trailing abort - with three sites
fixed by hand that the transform would have broken: a stub macro, and two
if/else-then-shared-abort blocks (PersistentMapTracker's adopt-tier and
SurfaceOpCodec's three surface-error branches) where a naive rewrite stole the
abort from the earlier branches and would have let them fall through. The
death tests are the safety net for byte-identity, and every one stayed green.

Four std::abort() remain under MG_Remote/, all inside the two funnels
(WireLog.cpp, FatalFunnel.cpp); the census exempts both files because their
callers carry the family word, and its baseline drops from 93 aborts to 4.
SessionFaultCount() is the one telemetry point S8 will assert zero over a good
run.

G1 re-measured: 0 symbol diff, .text unchanged at 0xa52203 (FatalFunnel and
FatalFamily live under MG_Remote/, compiled only in the disaggregated build).
Unit 2358 (one pre-existing environmental failure), integration-split 186/186,
integration-spawn 102/102, retrace 6/6, census 0 unmarked.
2026-09-21 21:44:17 -04:00
swung0x48 3a12a48add [MG_Remote] (P6/dl): the Fatal-family vocabulary as a table, projected onto the wire's seven codes
CONTRACT-P6 5.2's D2 ruling, step one of the Session::Fail funnel: the internal
vocabulary before the funnel that routes through it.

a6 censused thirty family words against a seven-value wire FatalCode, so
Session::Fail(FatalCode, detail) cannot carry the vocabulary in its first
argument - the word would collapse to one of seven and the log would say
ProtocolCorruption where the code said UnmigratedVerb. FatalFamilies.def is the
internal enum; the wire keeps FatalCode; each row carries the coarse code the
PEER is told when that family ends the session.

THE PROJECTION IS A TOTAL TABLE, NOT A SWITCH WITH A DEFAULT. This build has no
-Werror anywhere, so a default arm would silently swallow a family added
without a code. A row IS the arm; a family with no row is not a value of the
enum. FatalFamily.h generates the enum, the projection and the name from the
one .def, so a family, its word and its code cannot drift apart.

The buckets are documented per row in the .def: ProtocolCorruption for
malformed records / unimplemented verbs / role invariants / reply-contract
breaks; RingOverrun for a ring or event channel overrun; SegmentMismatch for a
segment too small for what a role must place in it; AbiMismatch for layout or
version disagreement; ServerCrashed for the liveness ends a WAITING peer
attributes to the other side. DeviceLost gets no row on purpose - it is armed
from the doorbell (5.3), never raised as a Fatal, and a test asserts no family
projects onto it so an abort can never masquerade as a device loss.

Five tests, all walking the table so a new row is covered without editing them:
count agreement, every family projects to a non-None code, none projects onto
DeviceLost, names distinct and non-empty, and a handful of anchors pinned by
meaning so a bulk-edit drift is caught.

No call site changes yet; the funnel that uses this is the next step.
2026-09-21 21:33:05 -04:00
swung0x48 5fb54f03fe [MG_Util, MG_Test, MG_IntegrationTest, CI] (P6): one log per role, and every reader moved with it
The spawn work had the two roles append to one MOBILEGL_LOG_FILE_PATH under
O_APPEND. It was correct - two processes at one offset would corrupt the file,
O_APPEND made each write atomic - but it put two sessions' lines in one place
and made every per-side assertion a search. This splits the sink: the client
role writes <base>.client.log and the server role writes <base>.server.log.

THE BASE NAME NAMES NO FILE ANY MORE, AND THAT IS THE POINT. Leaving the
client at the old unsuffixed path would have been the compatible choice and the
wrong one: a reader that was not updated would go on finding a file, go on
passing, and go on seeing only half the session - and a refusal census that
reports a confident zero for the half it cannot see is worse than one that
fails. With both names moved, an un-updated reader gets "file not found" and
says so. There were 34 such readers; every one is moved here.

THE ROLE IS PER-THREAD, not per-process, because under inproc both roles are
threads of one process. Log.cpp keys the sink on a thread-local role: the apply
thread declares itself server (ServerLoop.cpp), a spawned image is server for
its whole life (MOBILEGL_IPC_ROLE=server, read once), and everything else is
the client. The naming rule (RoleLogPath) and the two whole-log operations
(ReadRoleLogs / TruncateRoleLogs) are EXPORTED from the library, so no reader
reimplements the derivation - a second copy is the copy that falls behind.

Readers, by how they read:
 - unit mains set MOBILEGL_LOG_FILE_PATH and read a role file: they keep the
   base in the env and derive the client path from the library helper.
 - the death-test child runners (ReadLog) now concatenate BOTH roles: a death
   test asserts the CHILD said something, and the applier's refusals - the very
   diagnostics these check - are written under the server role.
 - the offset-based Emit readers derive the client path once after setenv.
 - PipeStatsWindow reads both, because draws/vbs increment in the backend's
   PrepareForDraw - the apply thread, the server role - so the summary line is
   in the server file even under inproc.
 - run_trace_case.cmake's split census reads both files; the client keeps the
   transport markers.
 - split_log_paths.py (and its self-test, and the control stub) derive both
   role files from the declared base.
 - the Android script pulls both role files off the device and concatenates
   them, so every downstream check is unchanged - and it judges a role file by
   CONTENT, because `run-as test -f` through `exec-out` returned 0 for a missing
   file on the device and `exec-out` folds cat's own not-found line into stdout.
   The monolith arm legitimately has no server file, and that is a silent skip
   rather than a one-line server log.

Guarded so a pull build is untouched: RoleLogPath and the role sink are wholly
inside MOBILEGL_BUILD_DISAGGREGATED; a non-disaggregated build writes the base
path exactly as before. G1 re-measured - 0 symbol diff, .text unchanged at
0xa52203.

Verified on device 2f7cbe2e: monolith writes only the client log (no phantom
server log), inproc and spawn each write both and pass their transport proofs
(spawn: server role in pid 16125). Unit 2353 (one pre-existing environmental
failure), integration-split 186/186, integration-spawn 102/102, retrace 6/6,
control smoke 12/12, split-log-paths self-test 9/9.
2026-09-21 21:26:05 -04:00
swung0x48 15f5b54bbd [MG_IntegrationTest, CI] (P6/t6): integration-spawn, and the set identity made structural instead of maintained
Exit gates 9.2 and 9.3. Until now nothing ran the integration suite across a
process boundary at all - the spawn shape was proven only by the retrace lane,
which exercises one trace per case and no scenario assertions.

integration-spawn: 102 cases, 102 green.

THE SET IDENTITY IS THE POINT, AND IT IS NOT MAINTAINED BY HAND. 9.2 asks that
the spawn lane's case-name set equal the split lane's. Registering the spawn
arm as a second copy of each block would satisfy that today and drift the first
time somebody adds a scenario to one of them - and a spawn lane covering LESS
goes green while proving less, with nothing about a smaller green lane looking
wrong. The eleven split-environment blocks become eleven calls to one macro
that emits BOTH arms from one filter, so a case added to either is added to
both or to neither.

scripts/ci/spawn_lane_parity.py asserts it anyway, because a macro is one edit
from a hand-written block. It compares SETS rather than counts - a count can
agree while the membership does not - and carries exactly one sanctioned
asymmetry, named in the file: integration-split also labels the PUSH-MONOLITH
arm, whose scenario is monolith-only by construction. R-16: introduce a
split-only block and it reds naming the case and the macro to use.

THREE CASES SKIP UNDER SPAWN, BY NAME. CtWireScenario reads PipeApplier's
in-process tallies through ServerVerbs(); under spawn those live in the SERVER's
process and this one's copy is 0 forever, so the assertions were failing about
somebody else's memory. That is a harness limit, not a product defect - the
records themselves cross, which the retrace lane proves. They are SKIPPED and
not deregistered: a case dropped from the registration would make the two sets
differ silently, which is the drift the macro exists to prevent. What retires
the skip is cross-process telemetry, which is also what S8's SessionFaultCount
needs.

TWO REAL REGRESSIONS THE EXISTING GATES CAUGHT, both from this refactor:

  * the macro dropped the per-block TEST_LIST variables, which were what
    SplitLogPaths fed to give every entry a PRIVATE MOBILEGL_LOG_FILE_PATH.
    SplitLogPaths.PrivateAndDistinct went red naming the first orphan. The
    macro accumulates list NAMES now, so a block added later is covered without
    anyone remembering to extend a list - and the names have to be unique,
    because gtest_discover_tests' TEST_LIST is populated when CTEST runs, not
    at configure time, so one shared variable held only the last block's
    entries. That matters more on the spawn arm: there TWO PROCESSES append to
    the path, so a shared file interleaves two sessions rather than merely
    being ambiguous.

  * the configure_file that substitutes those names ran BEFORE the F1 and Ct
    blocks registered, so it captured neither. Same gate, naming
    F1WireScenario.ClearBufferfiPixels. It now runs last.

junit_tally gains --require-spawn-ran beside --require-split-ran: the prefix IS
the arm, and a shared flag would let one lane tally the other's entries and call
itself run. Verified to discriminate - the spawn junit reads 96/0/5 under its
own flag and 0/0/0 under the split flag, which correctly fails.

Two death tests follow the ADOPT_TIER message the census commit retagged with
Fatal{UnimplementedAdoptTier}; their regexes asserted the old wording.

Unit 2353 (one pre-existing environmental failure), integration-split 186/186,
integration-spawn 102/102, retrace 6/6, fatal census 0 unmarked, G1 0 symbols
and .text unchanged at 0xa52203.
2026-09-21 20:20:24 -04:00
swung0x48 3399b8791c [MG_Remote, CI] (P6/dl): the Fatal census gate, and the one death under MG_Remote that named nothing
Exit gate S7, and the precondition for CONTRACT-P6 5.2's funnel rather than
its replacement.

S7 says "a bare abort() added outside Session::Fail -> the census gate goes
red". Session::Fail does not exist: it is the 93-site refactor 5.2 describes,
and it is not in this commit. What IS checkable today is the property that
funnel exists to guarantee and that every consumer already depends on - every
death under MG_Remote/ names a FAMILY, in a Fatal{Word} marker a grep can
find. run_trace_case.cmake reds a split retrace on any `Fatal{` line, the
retrace lane's refusal census counts them by name, and MEASUREMENTS quotes
the distinct words. A death with no word is invisible to all three.

TWO RULES, BECAUSE ONE WOULD LAUNDER THE OTHER. An abort must carry a marker
within twelve lines; and every WireLogFatal call must carry one in its format
string. WireLog.cpp's own abort is exempt from the first rule ONLY because its
five callers satisfy the second - a funnel that accepted an unmarked string
would be a hole shaped exactly like the rule.

THE BASELINE BOUNDS SILENCE, NOT DEATHS. Abort sites may grow freely; a new
family word may not appear unrecorded. Freezing the total would make every new
guard a gate change, which is how a gate stops being run. The wire's FatalCode
has seven values and the internal vocabulary is thirty-two, which is the
divergence D2 exists to bound - so growing it has to be an act, not an
accident.

THE CENSUS ALSO CORRECTS ITSELF. Its first run reported WireLog.h's own PROSE
as an unmarked death: the comment explaining why the stderr echo exists
contains the words std::abort(). So the scanner strips comments before
matching, with a character-level pass rather than a regex - a `"http://"`
inside a string literal is not a comment - while KEEPING string literals,
because the markers it looks for live in them. Real numbers after that fix:
93 abort sites, 32 family words, 0 unmarked. a6 had counted 92 and 30 from a
scan that did not make the distinction.

AND THE ONE REAL FINDING. a6 named two aborts with no marker at all;
WireLog.cpp's is the sanctioned funnel, so the genuine one is
PersistentMapTracker's invalid-MOBILEGL_IPC_ADOPT_TIER death. It now carries
Fatal{UnimplementedAdoptTier}, which is what makes the census zero.

R-16 both ways, run: a bare std::abort() reds it naming the file and line, and
a new family word reds it naming the word and how to record it.
2026-09-21 19:42:18 -04:00
swung0x48 e3bb6d8b82 [MG_Pipe, MG_Backend, MG_Impl] (P6/sm, P6/st): four guards that were compiled in and could never arm, and the crash behind one of them
CONTRACT-P6 3.2 (D1c), 3.3 (D13) and 3.4 (D10).

Every guard below asked a question a SERVER PROCESS answers wrongly - "is a
ClientSession active" (never, there) or "am I on the apply thread" (every
thread is, there) - so all four were present, compiled, and permanently
disarmed in the one shape they exist for. That is worse than absent: it reads
as coverage.

D13 FIRST, THEN THE WIDENING. MGPipeRoleSplitActive() meant "the REHEARSAL
knob is on AND the transport is real", so widening it in place would have left
a narrow name describing a wide predicate. It is renamed to
MGPipeRoleSplitRehearsalActive() across its 12 sites - including the
DualBlockScenario assertion message that embeds it - and the wider predicate
is introduced BESIDE it.

MGPipeBlocksAreDistinct() IS TRUE FOR TWO DIFFERENT REASONS, which is the
whole point: under the rehearsal the two blocks are distinct objects in one
process; under SPAWN they are distinct because they are in different ADDRESS
SPACES, and no knob is involved. The spawn retrace runs with
role-split-state=0 - measured off the lane's own log - so every guard gated on
the rehearsal was off there.

THE CRASH. MGPipeServerBlockNoteIdentity early-returned on the rehearsal, so a
spawn server's gPipeInputs.ContextIdentity() stayed nullptr for the life of the
process. DirectGLES' framebuffer binding-slot memo compares that identity FIRST
and with NO generation, against a cache variable that also starts nullptr - so
nullptr == nullptr reads as a CACHE HIT and the function returns a slot pointer
nothing ever filled. `sm` widens the gate; `st` lands the named Fatal
PipeInputs.h had already asked for, so a reopening says which memo and which
identity instead of corrupting memory three frames later. Guarded on
MOBILEGL_PIPE_PUSH: in a pull build that identity is the live GLContext's
address and is never null, so the test would be dead code and G1 measures that.

DISJUNCTS, NOT SUBSTITUTIONS - and this was measured the hard way. The first
attempt REPLACED OnApplyThread() with MGPipeServerArm() and
ClientSession::Active() with MGPipeSessionLive(). 14 RemoteGuards and
RemoteClientControls cases went green by omission: a ServerLoop fixture with no
client session has only the original facts, so swapping them out disarmed every
test that drives the apply thread directly. The originals stay; the new
predicates are added beside them.

AND THE PROBE HAD TO BE THE RIGHT ONE. Installing
ClientSession::ApplyThreadIsInsideApplier as the apply-thread probe made every
GL-thread call a role violation whenever the applier happened to be busy - the
inproc retrace died on Fatal{RoleViolation, "MGPipeSlots"} raised from the GL
thread. "Am I the apply thread" and "is the apply thread inside the applier"
are different questions; the second is R-1's mutual-exclusion probe and is
about another thread entirely. ServerLoop::OnApplyThread is the one the guards
mean.

Three of D1c's six predicates land here, the three the four guards need. The
other three (MGPipeSplitActive, MGPipeShouldDial, MGPipeBackendIsLocal) have no
D10 site and are not added as dead code; their call sites are ContextEpoch and
ActiveXfbState, which this commit does not touch.

D14 holds: the non-disaggregated build gets constexpr false definitions, not
runtime-false predicates, so every consult folds at compile time. G1
re-measured - 0 symbol diff, .text unchanged at 0xa52203.

Two test names added, both falsifiable by construction: the spawn case asserts
the blocks are distinct with the rehearsal knob OFF, and the monolith case
asserts the predicate is not simply a constant. Unit suite 2353 tests, one
pre-existing environmental failure. All six OpenRA arms green.
2026-09-21 19:38:19 -04:00
swung0x48 90f9a79058 [MG_Remote] (P6/hs): the build stamp that was silently empty, and two asymmetries the handshake never checked
CONTRACT-P6 4.2 and 4.3. Three defects, all of them things the handshake
believed it was checking and was not.

1. THE BUILD STAMP WAS "" AND NOTHING SAID SO. CMakeLists took the git hash
   through an execute_process with NO RESULT_VARIABLE, so a failed
   `git rev-parse` yielded the empty string. Verified live rather than
   inferred: this worktree's `.git` is a file holding a WINDOWS path, WSL's
   git answers `fatal: not a git repository`, result 128 - and every WSL build
   mixed "" into its ABI fingerprint. Two builds from different commits then
   agreed exactly, which is the one thing the stamp exists to prevent. The
   already-failing SessionHandshakeTest was reporting this the whole time.

   The fix may NOT touch GIT_COMMIT_HASH_SHORT: it is baked into
   glGetString(GL_VERSION), so redefining it changes an application-visible GL
   string and the pull build's .rodata - a behaviour leak and a G1 break at
   once. MGGitHash.h.in gains a SECOND pair instead, consumed only by the
   fingerprint, plus -DMOBILEGL_BUILD_STAMP for trees git cannot read (CI has
   ${GITHUB_SHA} in hand). A stampless build is now a configure-time WARNING
   naming the consequence, not silence.

   MOBILEGL_BUILD_STAMP_PRESENT IS MIXED AS ITS OWN FIELD. "this build cannot
   name its commit" and "this build's commit is the empty string" are different
   claims, and collapsing them is how every stampless build in the world agreed
   with every other. The test asserts the flag moves the fingerprint in EITHER
   build state, and the `BuildStamp = ""` perturbation is now guarded - in a
   stampless build production's value already IS "", so that assertion was
   passing while proving nothing.

2. THE CLIENT NEVER READ abiMajor/abiMinor. Hello has carried them since P5
   and the SERVER checks them; the client compared only the fingerprint. The
   fingerprint mixes AbiVersion so a mismatch would fire anyway - but as
   "struct shapes", which sends the reader looking for a layout change that is
   not there. Under spawn the peer is a SEPARATE IMAGE the operator chose with
   MOBILEGL_IPC_SERVER_PATH, so this is exactly the failure that needs its own
   sentence. Both handshake sites call one helper; there are two sites, and
   only one of them was reachable before `sm`, which is how the gap survived.

3. Hello::pid WAS 0 AND Welcome::serverPid ECHOED IT. So both were always 0 and
   neither named anything, while §9.5's arm-proof gate wants exactly the child
   pid. Both now carry real ids, and the pair is the cheapest arm proof in the
   protocol: under spawn they MUST differ, and a same-process session cannot
   produce two different values. The client keeps what the server STATED -
   local launcher knowledge is a different fact from one that crossed the wire,
   and only the second can be asserted from the server's own mouth.

R-16 both ways: revert serverPid to echoing hello->pid() and the new test reds
on both halves, naming 4.3. G1 re-measured after the CMake and header changes:
0 symbol diff, .text unchanged at 0xa52203.

Unit suite goes from two long-standing failures to one - the remaining
PipeCatalogue.FrontendNeverTakesAnApplierAddress is an environmental path
problem in this tree, unrelated. All six OpenRA arms still green.
2026-09-21 19:21:43 -04:00
swung0x48 8b13d8a84c [MG_Remote, MG_Impl] (P6/dl): the device-lost latch, taken from a descriptor and never from a deadline
Before this, CONTRACT-P6 5.3's own sentence held: "a dead server leaves the
client silently DECLINE-ing every verb forever with glGetError clean".
FatalCode::DeviceLost had zero producers and zero consumers and
glGetGraphicsResetStatus returned GL_NO_ERROR unconditionally, so an
application had no way to learn that the process holding its GPU objects was
gone. Measured cost of that: a server killed on its first Clear left the
client waiting out the full 120 s barrier and reporting Fatal{BarrierTimeout}
- the wrong diagnosis, two minutes late. It now notices in 0.01 s.

THE MACHINERY WAS ALREADY THERE. Doorbell::Wait breaks out on Dead(),
SessionProducer::Park turns that into SessionWait::ShutDown, and ClientSession
already had the named DECLINED path. What did not exist was the death FACT:
D5c's finding is that the client parks on clientBell[0] and rings ITSELF
through clientBell[1], holding both ends, so EOF can never arrive on the one
descriptor it watches. A bell that can wake itself cannot hear a death.

SetDeathWitness(fd) fixes that with no new descriptor: the control socket's
far end is held only by the server. The subtlety is that the bell must watch a
socket it must NEVER read - the transport's reassembler owns those bytes and a
control reply is not a death. So the witness enters the poll set with
`events = POLLRDHUP` and nothing else: POLLHUP/POLLERR/POLLNVAL arrive in
revents unrequested, so the hangup is seen while POLLIN never is. Nothing is
recv'd from it and its ownership stays with SocketTransport.

Dead() AND PeerHungUp() ARE TWO FACTS. Park sets m_dead for POLLERR, POLLNVAL
and unrecognised revents too - each a fault in THIS process's descriptor - and
CondVarDoorbell::Kill() sets it on every orderly Stop(). A latch taken from
Dead() would report a lost GPU for a local fd bug and a context reset for
every clean exit, so only PeerHungUp() arms it.

GL_UNKNOWN_CONTEXT_RESET, not GUILTY or INNOCENT: those assign blame, and the
server died for a reason that never crossed the wire.

D5b, in the control-reply wait, which had the right comment and no code: EOF
on the control stream is device loss; a TIMEOUT with the socket still open is
"ALIVE BUT SILENT" and says so by name. 5.4 forbids calling a frozen or merely
slow peer dead, and under P5e run-ahead a server one frame behind is the
INTENDED steady state - a latch armed by a deadline would fire on a healthy
session under load.

MOBILEGL_IPC_RESPAWN is now a NAMED REFUSAL rather than a variable nothing
reads. The latch is one-way on purpose: every handle this client minted names
an object inside the server process, so recovery means re-pushing the world.

THREE TEST NAMES ADDED, none renamed (G14), and R-16 run both ways:
  * remove the witness -> AKilledServerLatchesDeviceLostFromTheHangupAndNot-
    ADeadline goes red, takes 2 s instead of 0.01 s, and names D5c.
  * collapse PeerHungUp() into Dead() -> ADeadBellIsNotAlwaysAHungUpPeer goes
    red naming the local-fd-bug consequence.
  * AnOrderlyStopIsNotADeviceLoss stays GREEN under that collapse, and its
    comment now says so. The first draft claimed it was the control for it;
    running the falsification proved otherwise - under spawn the self bell is a
    SocketDoorbell and Stop() never kills one. The comment was corrected rather
    than the claim quietly kept.

G1 RE-MEASURED, NOT ASSUMED: 0 symbol diff against the c6 baseline and .text
unchanged at 0xa52203. That measurement earned its keep - it caught a 16-byte
drift that had nothing to do with dl: P6's O_APPEND log-sink fix was compiling
into the PULL build, where a second process can never exist. Now guarded, and
.text is back to the baseline exactly.

Full unit suite: 2350 tests, the same two pre-existing environmental failures
and no others. All six OpenRA arms (2 backends x monolith/inproc/spawn) green.
2026-09-21 18:49:56 -04:00
swung0x48 d83fc6643a [CI, android-plugin] (P6): CI starts a second MobileGL process on a device
The Android retrace matrix gains a third lane, spawn-acceptance, beside
monolith-control and inproc-acceptance. Until now nothing in CI had ever
started a second MobileGL process on a device at all.

The apk.yml retrace job already carried a `transport` matrix dimension and
already branched on it, so the lane itself is three edits: a row in
github_transport_matrix, an elif in the retrace step, and a summary to render.
What needed thought was the gate.

THE SPAWN LANE IS APK-ONLY, and that is a statement about where evidence is
missing rather than about what spawn can do. Linux already has
test.yml's `retrace-split`, which is {backend, case, transport} over inproc AND
spawn with the two negative controls that make it mean anything. A spawn row in
the Linux transport matrix would pay for a second full sweep proving nothing -
and test.yml's `retrace` job stages no server image, so the lane would not
merely be redundant, it would be red. self_test_transport_matrices now asserts
the absence in both directions, because an accidental removal of the `if apk`
has to be caught in the self-test rather than in CI.

--require-spawn REQUIRES TWO SENTENCES, and the second is the one that matters.
ConfigLoader's marker proves the VALUE RESOLVED - it exists only in the Spawn
arm, which exists only under MOBILEGL_BUILD_DISAGGREGATED - but that is a
statement about a parser. `spawn ARMED - the server role runs in pid N` is
written only after the launch, the connect and the handshake have all
succeeded, and the pid in it is the SERVER's. It is the only line in the log a
same-process run cannot produce.

Both halves red once, measured:
  * a monolith run demanded to prove spawn: REFUSED, transport_resolution_marker
    false, server_pid null - while the retrace itself scored ssim=1.000000,
    which is exactly the trap. OpenRA matches its golden under monolith too, so
    the picture has never been the evidence.
  * the same run's real spawn log with the one ARMED line deleted: REFUSED,
    transport_resolution_marker TRUE and server_pid null. That is the
    "resolved the transport and then never left the process" case, isolated.

THE PROOF USES `any` WHERE THE inproc GATE USES `all`, and the difference is
load-bearing rather than a relaxation. Under spawn two processes append to one
MOBILEGL_LOG_FILE_PATH, and the launcher deliberately scrubs every
MOBILEGL_IPC_* from the child's environment (anti-recursion catch (b)) - so the
server's own `Config: IPC` line reads strict=0 role-split-state=0 BY
CONSTRUCTION. Demanding every line match would red every spawn run for the one
thing the design requires.

--use-pbuffer is not optional on this lane: until P12 a spawned server cannot be
handed a window, because an ANativeWindow* is a pointer into the CLIENT's
process (Rule H). MOBILEGL_IPC_SERVER_PATH is deliberately NOT passed -
nativeLibraryDir carries an install-time hash that nothing off the device can
spell, and the Activity resolves it.

The local runner now passes the same knobs and the same --require-* flag CI
does, so "it works locally" can no longer mean "it ran monolith locally". That
change immediately caught its own omission: the first local run with the gate
armed failed on client_ipc_line_present=false because the runner had been
sending MOBILEGL_TRANSPORT alone.

Verified on device 2f7cbe2e: OpenRA.DirectGLES spawn, ssim=1.000000,
mismatchPixels=0, "real spawn transport, server role in pid 9885, and zero
Fatal records confirmed".
2026-09-21 18:30:18 -04:00
swung0x48 7c8120cc49 [MG_Remote, android-plugin] (P6): a real phone retraces a real trace across two processes
OpenRA.DirectGLES, transport=spawn, on a Redmi: ssim=1.000000,
mismatchPixels=0, zero Fatal{ lines, three runs for three. The server is a
process the client launched out of the APK's lib/arm64/, and the client's own
log names its pid.

Four things stood between the desktop green and the device green, and every
one of them was invisible on desktop for the same reason: a build tree is not
a deployment, and Linux is not Android.

1. THE SERVER COULD NOT LINK. It links libMobileGL.so and is EXEC'd, not
   dlopen'd, so it gets a fresh linker namespace with the DEFAULT search path
   - and an APK's lib/<abi>/ is not on it. `CANNOT LINK EXECUTABLE ...:
   library "libMobileGL.so" not found: needed by main executable`, before a
   single line of MobileGL ran. On desktop CMake had been giving the target an
   absolute build-tree RPATH, so the WSL server resolved its dependency
   through a path no phone has. $ORIGIN is asked for via BUILD_RPATH (a raw
   -Wl,-rpath,$ORIGIN reaches the generator unquoted and the SHELL eats it -
   the device linker said `normalize_path - invalid input: "$"`), and because
   the NDK drops RPATH from the link anyway, the LAUNCHER states
   LD_LIBRARY_PATH: it is the one component that has already resolved where
   the image lives, and a pair that is not co-located is not a pair.

2. THE RENDEZVOUS NEEDED A WRITABLE DIRECTORY. An Android app has no /tmp, so
   bind() in the child had nowhere to put the node and the client refused by
   name 20 s later. AF_UNIX endpoints now accept the ABSTRACT namespace with a
   leading '@' - the systemd/D-Bus convention - which has no filesystem
   presence at all: no directory, no mode bits, and no stale node to unlink
   after a crash, which is the bind() failure Listen() has been working
   around. Its reach is the network namespace, which for an app is the app.
   StartSpawned mints '@' names; an explicit path still works everywhere, and
   that is what P12's "connect to something somebody else is listening on"
   will use.

3. PBUFFER WAS A DirectGLES-ONLY KNOB. Twice over: trace-replay-ci.sh gated
   `--ez use_pbuffer true` on the backend name, and the JNI demanded a native
   window for DirectVulkan unconditionally. A surface SHAPE is not a backend
   property, and the coupling mattered because the Android GLWS makes a window
   surface whenever gNativeWindow is non-null - so handing Magma a window was
   the same as asking for one. Until P12 a spawned server cannot be given a
   window at all: an ANativeWindow* is a pointer into the CLIENT's process and
   SetWindowHandle is refused on the way across with Fatal{UnmigratedSurface,
   "AndroidNativeWindow@P12"} (Rule H). Which is exactly what happened, one op
   after the server had come up green.

4. THE RUNNER LOOKED IN ONE OF THE TWO OUTPUT DIRECTORIES. assembleTraceRelease
   writes apk/trace/release; CI has staged the same artifact under
   apk/trace/debug. A tree that has built locally has both, and picking the
   wrong one retraces last month's APK and says nothing about it. Newest of
   either, now.

MEASURED AND NOT FIXED: DirectVulkan on this device diverges under BOTH split
transports and does so nondeterministically - inproc scored ssim=0.988998 /
9836 px, spawn scored 0.976494 / 14658 px and then 0.988998 / 9836 px on a
repeat, while monolith with the same pbuffer surface scores 1.000000 / 0 px.
The two transports draw from the same set of outcomes, so this is the split
path on Magma against a real GPU, not the process boundary: P6 neither caused
it nor can fix it. DirectVulkan spawn IS green on desktop (lavapipe, 1.000000).

G14: one test name added, AnAbstractEndpointNeedsNoWritableDirectoryAndLeaves-
NoFile, which pins both halves - the handshake crosses an abstract name, and
nothing is left on the filesystem. A '@' mistaken for a path would have bound
a file of that literal name, passed here, and failed on the one platform this
exists for.
2026-09-21 13:33:57 -04:00
swung0x48 557dc2c109 [MG_Remote] (P6/dl): D5c - a bell that can wake itself cannot hear a death
Measured while running the 78-entry spawn matrix, not predicted.

minecraft-1.21.4-fabric-iris-iterationrp-in-world kills the SERVER process
outright on this machine - llvmpipe cannot lower the shader pack's
%llvm.x86.vcvtps2ph.256 and aborts at call 159429 - and the client did NOT
take the SessionWait::ShutDown path, which already exists and is already
correct. It sat out the full 120 s and reported Fatal{BarrierTimeout,
"ResourceRespecify"}: the wrong diagnosis, at two minutes a case, for a
server that had been gone since the first second. The same trace fails in 3 s
under monolith.

The cause is structural and is the price of `sm`'s two-fd bell. The client
parks on clientBell[0] and rings ITSELF through clientBell[1], so it holds
both ends of that socketpair. EOF arrives only when every writer closes and
the client is one of them, so that descriptor can never hang up no matter
what happens to the peer.

`dl` therefore needs a witness descriptor whose far end only the server
holds, polled beside the park fd and never read from, whose POLLHUP/EOF
latches Dead(). The control socket already is exactly that, so no new
descriptor is needed.

This is NOT a defect in `sm`: holding both ends is what self-notification
requires, and without it the server's control pump could not wake its own
apply thread - the bug that cost this branch an afternoon. The mistake would
be assuming one descriptor can answer both questions.

Recorded rather than fixed: it turns no red case green, and the only case
that reaches it fails identically in all three arms for a reason that is not
MobileGL's. `dl` owns the fix and S2 already names the red-once for it.
2026-09-21 12:42:03 -04:00
swung0x48 534067493c [MG_Remote, android-plugin] (P6): the device can run the split retrace, and only the app can spell the server's path
Two small pieces, so an Android spawn arm is a flag rather than a project.

THE RUNNER GAINS --transport. It already had a generic env passthrough - one
intent extra carries `K=V;K=V` all the way to setenv before the library is
dlopened - so the transport itself costs one entry and no new plumbing in the
five files between here and the device. monolith passes NOTHING rather than
MOBILEGL_TRANSPORT=monolith, so the default arm's environment stays
byte-identical to what it was before this option existed: a control that
changed when the option landed would not be one.

THE SERVER PATH CANNOT COME FROM THE HOST. Every other knob can, but this one
names a file that must be EXEC-ABLE, and on Android there is exactly one such
place: an APK's lib/<abi>/. W^X has forbidden the app's own data dirs since
API 29, and the packager only puts a file in lib/<abi>/ if it is named
lib*.so - which is why the server executable wears that name, and what P0
spike A proved on two devices. nativeLibraryDir contains an install-time
hash, so no host-side script can spell the path; the process about to launch
the server is the only one that can. TraceReplayActivity resolves it, and
only when the caller actually asked for spawn.

AN EXPLICIT MOBILEGL_IPC_SERVER_PATH STILL WINS. A caller that set one is
pointing somewhere deliberately - at a pushed build, say - and silently
replacing it would make that knob untestable on the one platform where it
matters most.

The APK must be built with -Pmobilegl.buildDisaggregated=ON or none of this
is reachable: without it ConfigLoader has no transport parser, accepts the
value, ignores it (CONTRACT-P5 rule 5), and the run is monolith under a name
that says spawn. The --transport help says so where somebody will read it.
2026-09-21 12:34:34 -04:00
swung0x48 dc5bd935de [MG_Remote, CI] (P6): the split retrace lane becomes the whole CI matrix, both backends, both transports
It was one case on one backend under one transport: OpenRA, DirectGLES,
inproc. It is now every CI case, both backends, inproc AND spawn - 154 CI
entries, 156 local ctest entries - because every defect the spawn work turned
up was a question about the SERVER's process and not about any one trace or
either backend.

THE DirectVulkan ARM. tools/trace_replay/CMakeLists.txt carried "a
DirectVulkan arm would be measuring a server nobody has written yet". True
when P5 wrote it; false since P5f/P7 landed Magma's server consumers and its
run-ahead readiness gate. Leaving it would have shipped a split shape whose
second backend nothing ever ran - the same class of hole as a lane that runs
monolith under a name that says split. Magma now brings up a server process
of its own and OpenRA retraces across it at ssim=1.000000, which is only
possible because this branch also taught the child to load its config: before
that, a DirectVulkan session brought up a DirectGLES server and the two
halves disagreed about which backend the capability mask describes.

`split` BECOMES AN OPT-OUT. It began as an opt-in naming one case, because
one case was the work. Marking 39 cases `"split": true` would carry no
information and would reinstate the hole the key was written to close, only
mirrored: a case added later would get the monolith arms and be silently
absent from the split ones, and absent is the one state nothing reds on. A CI
case is now in the subset unless it says `"split": false`, which is a
reviewable claim that a trace cannot cross a role boundary.

THE CI MATRIX GAINS A THIRD DIMENSION. The job took {backend, case} and
hard-coded MOBILEGL_TRANSPORT=inproc in its own run block, so `spawn` had no
way into CI at all; the .SPAWN ctest entries existed only for a local build.
It is {backend, case, transport} now, the backends are the case's own
ci_backends rather than a hard-coded DirectGLES, and the artifact names carry
the transport or the two arms of one {backend, case} would overwrite each
other's uploads.

inproc STAYS AS THE CONTROL. Same library, same codec, same applier, same
picture; the only difference is which process the server role runs in.
Running only spawn would leave "did the transport break it?" and "did the
phase break it?" answerable only together - which is the question every
failure in this phase has turned on, including the SIGSEGV that D1 explained.

BOTH NEGATIVE CONTROLS FOLLOW THE ARM. retrace_pull_library_control.sh and
retrace_drop_draw_control.sh each exported MOBILEGL_TRANSPORT=inproc
themselves, so on a spawn job they would have gone on proving something about
the other arm while spawn ran unguarded. They read SPLIT_TRANSPORT now, and
default to inproc so a caller that sets nothing behaves exactly as before.
scripts/ci/control_smoke_test.sh: 22 passed, 0 failed against the stubbed
ctest, including the two modes that matter most - a red whose SSIM never fell
and a knob that armed and dropped nothing.

The spawn arm needs a second binary and the inproc arm needs none, so the job
places libMobileGLServer.so only on the spawn arm and a missing one is a
named failure there rather than a LaunchServer refusal 78 jobs deep.

Measured locally as this lands: 39 of the 78 SPAWN entries complete, all
green. The rest are still running.
2026-09-21 12:31:49 -04:00
swung0x48 7e244b3775 [MG_Remote] (P6/cp): a real trace retraces across two processes, pixel-for-pixel
OpenRA.DirectGLES.SPAWN is green: ssim=1.000000, mismatchPixels=0, zero
Fatal{ lines, against a server in a process the client started by name. The
inproc SPLIT arm stays beside it as the control, so "did the transport break
it" and "did the phase break it" remain separately answerable.

Five defects stood between the handshake and that picture. Each one was
silent in its own way, and none of them looked like what it was.

1. THE DOORBELL COULD NOT WAKE ITSELF. SocketDoorbell's one-fd form sends on
   the end it polls, so a write is delivered to the PEER - correct for waking
   the other process, useless for waking a thread in this one. The server's
   control pump posts into the apply thread's mailbox and must ring the bell
   that thread parks on. Symptom: every eglInitialize crossed, landed in the
   mailbox, and timed out at 5 s. The two-fd form (park on [0], notify on [1])
   is restored here, and the bell handover goes from two descriptors to THREE
   - the client needs the same split for the same reason. Seven offers now.

2. InitCapabilities HAD NO WIRE KIND. CONTRACT-P6 §2 predicted this exactly
   ("the request has a live client caller and the codec refuses to encode
   it"). SurfaceOpKind gains 11, append-only; the decode range's upper bound
   moves with it, or the new tag reads as out-of-range and surfaces as
   Fatal{ProtocolCorruption} rather than the missing row it is. The op's name
   loses its InprocOnly suffix, which stopped being true here.

3. THE CHILD LOADED NO CONFIG AT ALL. mobilegl_server_main calls
   InitServerRoleForSpawn directly, and MG_ConfigLoader::Init is a step of
   MobileGL::Initialize - so every MG_Config value in the server was its
   static default. MOBILEGL_BACKEND_TYPE survives the launcher's scrub on
   purpose and was read by nobody: a DirectVulkan session would have brought
   up a DirectGLES server and then disagreed with its client about which
   backend the capability mask describes.

4. AND THEREFORE Transport READ AS Monolith, which CONTRACT-P6 §3.1 (D1)
   forbids in as many words. a6 measured the cost: 226 live lines test
   `Transport != Monolith` to select the SERVER arm, 148 in MG_Backend.
   Monolith flips all of them to frontend glue in the one process with no
   frontend. Observed precisely as predicted - UnitTexturesByHandle() answered
   false, Clear took the legacy arm that dereferences borrowed FRONTEND slot
   pointers, and the apply thread took a SIGSEGV inside SyncNeccessaryTextures
   on the first Clear record of the trace. It is pinned in-process rather than
   through the environment because the environment is where it must NOT be:
   LooksLikeAClientEnvironment exits 65 on a surviving MOBILEGL_TRANSPORT, and
   that check is anti-recursion catch (b). Dial, not Transport, is that axis.

5. THE LOG WAS CORRUPT, NOT MERELY INTERLEAVED. Two processes opened the same
   MOBILEGL_LOG_FILE_PATH with "w" and each carried its OWN file offset, so
   they overwrote each other's bytes - half-lines like `ote client: CapsMirror
   generation 1 adopted`, and the server's whole startup block vanishing under
   the client's later writes. This cost real diagnosis time twice, because a
   missing line read as a code path not taken. O_APPEND now; the client still
   truncates once so a fresh run is not a pile of old ones, and the server
   never does - it would delete the very startup lines a failed session needs.
   run_trace_case.cmake's Fatal{ census reads exactly this file, so it was
   also reading a lie.

The retrace runner's split assertions become transport-aware: one row per
transport it admits, and a transport with no row is a FATAL rather than a
fall-through, because the fall-through is the accident the block exists to
prevent. The spawn row carries a SECOND sentence and it is the stronger half
- ConfigLoader's marker proves the value RESOLVED, while `spawn ARMED - the
server role runs in pid N` is written only after the launch, the connect and
the handshake all succeeded, and carries a pid that is not this process's.
inproc resolves its own marker while running the server role on a thread
here; nothing on that path can write this one.

G14: one test name added (TheEleventhFramedOpIsInitCapabilitiesAndItCrosses-
TheWire), none removed. The ten-op table keeps its name and its scope rather
than being renamed to eleven.

Two unit failures predate this change and are environmental in this tree:
PipeCatalogue.FrontendNeverTakesAnApplierAddress (the frontend source root
resolves empty) and SessionHandshakeTest.TheAbiFingerprintChangesWhenAnyOf-
ItsInputsDoes (GIT_COMMIT_HASH_SHORT is empty here, which a6 already recorded
as red in this configuration).
2026-09-21 12:23:41 -04:00
swung0x48 5239ae0c19 [MG_Remote] (P6/cp): EGL control frames cross to the other process, and answer
An eglInitialize issued in one process is dispatched in another and its answer
comes back. That is the last thing standing between the session and a native
context, because the server creates its context on the apply thread when the
client's first eglMakeCurrent crosses - and until now nothing crossed.

ONE SEAM, NOT TWELVE. a6 row A4-13 counted twelve Server* forwarders, each
ending in ServerLoopInstance().RunSurfaceControlFrame(frame) with no role
branch, and concluded twelve edits. They all funnel through that one function,
so the branch goes there and the forwarders are untouched.

A HOOK, NOT A CALL. ServerLoop::SetRemoteControlSink takes a function pointer
the client installs at handshake. The obvious alternative - the server half
asking ClientSession::Active() - would have added a fifth symbol to the six
MG_Remote already owes itself (a6-link-experiment §5), in the one file with a
clean reason not to.

The branch is BEFORE the local seq mint, deliberately. RunSurfaceControlFrame
numbers unnumbered local posts; a wire request must keep the CLIENT's number all
the way through dispatch and reply or its answer cannot be correlated
(CONTRACT-P6 §6.3, which says the client mints, always).

The client sink encodes with EncodeSurfaceOpFrame, sends, and waits for the
SurfaceReply carrying its own seq - under one mutex, because the twelve
forwarders are already caller-serialised on the server's m_callerMutex under
inproc and two replies racing for one correlation is the same bug on either
side. A CapsSnapshot arriving while it waits is ADOPTED rather than dropped:
the server republishes caps from inside its MakeCurrent dispatch (a6 row A4-30)
so a snapshot legitimately precedes the reply to the op that produced it. The
wait is bounded and its expiry is NOT fatal - a peer that is gone is a different
fact from a peer that is slow, and only the doorbell's death latch may say which.

The server's pump decodes and applies through ServerApplyWireSurfaceOp, which
a6 found had no production caller anywhere. It runs on ServerMain's own thread -
NOT the apply thread - which is what §6.2 asks for: a control arrival must not
wake the applier, and frame decode must not happen on the thread holding the
native context. A frame that is not a verifiable SurfaceOp ends the session
rather than being ignored, because a silent drop is the failure mode this
project refuses everywhere else.

THE TEST'S ASSERTION IS THE OUT-PARAMETERS, not the EGL result. A headless
machine has no display to initialise, so the op is expected to be refused - and
ServerInitializeEGLDisplay writes major/minor "whenever the dispatch ran,
success or not" (the forwarder's own comment). They move off -1 if and only if
a SurfaceReply came back from the other process; a transport failure returns
before the write-back. I got this backwards first and the test said so.

Gates: ServerSpawnTest 5/5; unit suite 2345 with the same two pre-existing
environmental failures; G1 pull 0 added / 0 removed, .text 10822147 unchanged.

Not TCP. It was considered and dropped for now: TCP has no SCM_RIGHTS, so the
four segments cannot cross it, and a TCP control plane would have to refuse a
data plane by name until StreamLink exists (P6.5). Recorded here so the reason
is on file rather than re-derived.
2026-09-21 10:50:28 -04:00
swung0x48 11807ac7d0 [MG_Remote] (P6/sm): two INDEPENDENT processes, joined by a name, sharing memory
The cross-process session is up. A server process is started on its own with an
endpoint and nothing else, it listens; a client started separately connects to
that name; Hello crosses, Welcome comes back with four SegmentRefs, six
descriptors cross by SCM_RIGHTS, and the client maps memory another process
created. ClientSession::Active() is non-null at the end of it, which is the flag
that arms the thirty-seven wire emitters.

INDEPENDENT, NOT FORKED - and that is a correction to this package's first
shape. The version committed an hour ago forked the server and handed it fds
3..6. It worked, and it was less code, but inheritance can only ever produce a
server that is a CHILD OF ITS CLIENT, which is the one arrangement the end state
cannot use: there the server is a standing application and the client is
somewhere else entirely, possibly on another kernel. CONTRACT-P6 §3.1 called
this Dial == Connect and filed it as a later shape. It is the primary one, and
fork/execve survives only as a way to START a process - every descriptor it used
to carry is gone.

What that costs and buys, concretely: the rendezvous becomes a NAME, so
SocketTransport grows Listen/AcceptPair/ConnectTo, with a bounded connect retry
that is the CLIENT's (a launcher that waited for the server to be ready would be
coupling the two again by the back door). The bells stop being inherited fds and
are created by the server and handed over with the segments - slots 4 and 5 of
six SCM_RIGHTS offers. And the client connects TWICE to the same path: an fd
offer is a sendmsg whose ancillary data rides with specific bytes, so it may not
share a socket with the framed control stream, or the frame reassembler and the
descriptor receiver race for the same bytes.

The pieces behind it:
  - SessionSegments::AdoptFromDescriptors, which is the same Adopt+Map loop
    AttachInProcess has run on every inproc session since P5. That was the point
    of writing the inproc attach as dup+Adopt+Map rather than as an alias: only
    the origin of the descriptor is new.
  - ServerSession::SetExternalDoorbells, the narrow fix for a6 row A7-1 - the
    §3.9 accessors abort for every role but InProcess, so a spawn apply thread
    died before popping its first record. lk deletes them; until then an injected
    pair is honest about where the bells come from.
  - MG_Backend::InitServerRoleForSpawn: steps 1 and 2 of the split bring-up
    extracted so the server process runs the SAME code the inproc server role
    runs. The capability bits are a statement about the server's own backend and
    a second copy would diverge the first time a backend registers a slot.
  - ClientSession::FinishStartup: everything below the attach - producer, reply
    pool, event ring, segment table, encoder, first CapsSnapshot, emitter tables -
    shared verbatim between inproc and spawn. Its one parameter is not a
    transport question: under inproc the client starts the apply thread because
    there is no server-side main; under spawn there is one, and it started the
    thread before it sent the first descriptor.

G1 CAUGHT A REAL LEAK on the way: InitServerRoleForSpawn was guarded on the
inside, so the pull build still DEFINED the symbol - 2 added, 1 removed, .text
+16 bytes. The whole function and its declaration are now inside the guard.
"It returns false there" is not the same as "it is not there", and the gate is
what said so.

Gates: ServerSpawnTest 4/4; unit suite 2344 with the same two pre-existing
environmental failures; G1 pull 0 added / 0 removed, .text 10822147 unchanged.
2026-09-21 10:36:14 -04:00
swung0x48 70c45ab9ce [MG_Remote] (P6/sm): two MobileGL processes, talking
The first time this project has had a second process. fork + execve of a real
server image, the control socket surviving the exec, bytes crossing a process
boundary, and the child exiting on EOF with nothing left behind.

mobilegl_server_main EXISTS NOW. a6 found ARCHITECTURE.md:488-496 had described
it for three phases while the symbol was nowhere in the tree, and the only
MobileGLServer target was the P0 spike stub, which links no MobileGL library at
all. The spike keeps its evidence and its name becomes MobileGLServerSpike; the
real image links the library, so an undefined symbol is a build error rather
than a dlsym failure in a child whose stderr is /dev/null.

WHAT sm OWNS AND WHAT IT LEAVES: the process, the inherited descriptors, the
transport and the exit discipline. Not the session, not the applier, not the
four segments - wiring those from ServerMain before lk has a seam to wire them
THROUGH is how call sites get written twice. The loop is a read/answer/exit-on-
EOF stand-in whose SHAPE is the part that matters: one reader, EOF ends the
process, and no timeout is ever allowed to stand in for the death fact (§5.4).

ANTI-RECURSION IS TWO CATCHES, and neither is ARCHITECTURE.md:488's. That rule
says the child hard-sets Transport to Monolith; a6 measured that 226 live lines
test `Transport != Monolith` to select the SERVER arm, 148 in MG_Backend, so
obeying it flips the whole backend to frontend glue in the one process with no
frontend. Instead: (a) MOBILEGL_IPC_DIAL=no, structural and separate from role,
and (b) MOBILEGL_TRANSPORT and every MOBILEGL_IPC_* scrubbed from the child's
envp. The child refuses with 64 for (a) and 65 for (b) so the two safeties are
distinguishable - two that cannot be told apart are one.

TWO BUGS WORTH RECORDING, both found by the test rather than by review:

dup2(n, n) IS A NO-OP THAT DOES NOT CLEAR CLOSE-ON-EXEC. FdPassing sets
SOCK_CLOEXEC on the aux pair, so the child depends on dup2 to clear it - but in
a gtest process a fresh socketpair lands on exactly fds 3..8, so a source fd
often IS its own target. The no-op left CLOEXEC set and the following close()
then shut the descriptor we had just "installed". The child came up with no
socket and the parent saw TRANSPORT_CLOSED on its first send. Now spelled out
rather than written cleverly.

THE CHILD'S DIAGNOSTICS WERE INVISIBLE. WireLogError reaches the file sink
only - Defines.h builds the logger with the console sink off - so a child that
refused to start said so into a file the parent does not know the path of.
ARCHITECTURE.md §15.2 warns about exactly this on Android. Both refusal paths
now also write stderr directly, and both carry a distinct exit code the parent
can name without reading anything.

TESTS: six cases, 6/6. Each exists because something would otherwise pass
silently - the reply carries the child's own getpid() so a test that talked to
itself cannot pass (ID-124's lesson at the smallest scale); the process tree is
COUNTED before, during and after, not promised; the child is proven not to exit
while merely idle before it is proven to exit on EOF; an unresolvable image is a
named refusal that spawns nothing (S1); a surviving client knob is caught by
name (S3); and a descriptor crosses execve by SCM_RIGHTS, which is the mechanism
the four segments will ride on and is now exercised before it is needed.

Also records the asio decision in the contract: declined for P6 because spawn
needs five platform mechanisms and asio covers one - its own socketpair returns
operation_not_supported on Windows, it has no descriptor passing, segment
delivery is POSIX-only by design, and there is no fork. Revisit at P6.5, where
asio::generic::stream_protocol carries a raw family/type/protocol triple and
therefore gives AF_VSOCK from one implementation.

Gates: unit suite 2346, up 6, same two pre-existing environmental failures. G1
pull 0 added / 0 removed, .text size 10822147 unchanged, and zero spawn, server
or SocketTransport symbols in the pull build.
2026-09-21 10:04:31 -04:00
swung0x48 5f846d96e3 [MG_Remote] (P6/so): SocketTransport, green over a real socketpair before any second process
The first P6 code that runs. ITransport over a connected AF_UNIX SOCK_STREAM
socket, with SCM_RIGHTS descriptor passing on a dedicated SOCK_DGRAM aux pair.
No shared memory: this is the CONTROL plane, and the hot path bypasses
ITransport entirely by its own header's account. What the socket carries is the
handshake, surface ops, resync, aux requests and fatals - and, once sm exists,
the four segment descriptors that let the peer map the data plane. It moves the
keys, not the bytes.

CONTRACT-P6 §8 and the spawn plan both require this green over a plain
socketpair with two threads BEFORE a child is ever forked, for the reason
SessionRings.h gives about the attach half: a transport first exercised on the
day the process appears is a transport whose bugs all arrive at once, on the
hardest day to debug them. That requirement is now met.

It reinvents nothing. Framing.h already had AppendFrame and a real byte-stream
reassembler with exactly the BUFFER_TOO_SMALL keep-the-message semantics
ITransport specifies - written for this and never before driven by a socket.
FdPassing.h already had SCM_RIGHTS over a dedicated datagram pair. What is new
is ~250 lines joining them, plus the platform refusal: Windows has no
SCM_RIGHTS and ShmSegment::Adopt is POSIX-only, so CreatePair refuses BY NAME
at construction rather than handing back a transport that fails every later
call.

SOCK_STREAM rather than SOCK_SEQPACKET, deliberately: Doorbell.h:390-398
already chose a stream socket so a closed peer surfaces as POLLIN|POLLHUP with
recv()==0 rather than as a hang, and the control plane may not disagree with
the bell about what death looks like. Framing.h is what makes a stream safe -
the message boundary is in the payload, not in the socket.

TESTS: the same nine ITransport-contract cases InProcessTransportTest runs,
copied rather than shared. Folding both suites into one parameterised fixture
would rename the sixteen existing cases and G14 says names only grow; and the
two suites do not share a handle anyway, since InProcessTransport's tests reach
into condvar doorbells this class does not have. Nine names added, zero
removed. The seven Doorbell* cases are deliberately absent: they exercise
CondVarDoorbell, not ITransport.

Plus three a stream has and a message queue cannot: a 256 KiB payload that
provably crosses in several reads; a garbage stream that must latch failed and
must NOT resync onto a boundary it invented; and adopting an already-connected
fd with no aux channel, which answers UNSUPPORTED to ShareFd rather than
pretending - rule G, a transport declares what it supports.

Red-once, both run: drop the message on BUFFER_TOO_SMALL and six cases go red;
make the framing latch recoverable and AGarbageStreamIsLatchedFailed fails by
name.

Gates: 12/12 in 677 ms. Unit suite 2340 tests, up 12, with the same two
pre-existing environmental failures. G1 pull 0 added / 0 removed, .text size
10822147 unchanged, and zero SocketTransport symbols in the pull build.

One note for the VM track this eventually serves: swapping AF_UNIX for
AF_VSOCK is one line here and nothing at all in Framing. It is the shared-memory
half that does not cross a kernel.
2026-09-21 09:47:32 -04:00
swung0x48 db62d674d0 [MG_Remote] (P6/lk): regroup RingControl by writer, and make the asserts that pin it falsifiable
The shared control page both roles map had a mixed-writer cache line:
{appliedSeq, submittedSeq, retiredSeq, completedFrameSerial, presentAckSerial}
sat together, but submittedSeq is producer-written (Ring.h's "THE FIVE
WATERMARKS") while the other four are consumer-written. That is why no
single-writer progress block could be aliased over it, and it is what this
package fixes: submittedSeq moves up beside cmdHead, the other thing the
producer writes, and the four that remain become a named LinkProgress member -
the struct c6 declared in ILink.h - so a data plane with no shared page can own
the same four values and every wait predicate in SessionRings.h keeps ONE
implementation at source level.

The two event flags deliberately stay in the doorbell group. Promoting them
would put the client's per-drain exchange(0) on eventRingFull (EventRing.h:200)
onto the line the client's own spin predicate reads 1e5-1e6 times a frame.

73 call sites rewritten, 15 in the library and 58 in tests. The rewrite is
receiver-scoped rather than field-scoped, because SplitRuntimePeek.h:77 has its
own struct with a field of the same name and SplitRuntimePeek.cpp:146 has both
on one line - a field-scoped rename would have silently broken the peek struct
that MagmaRunAheadScenario reads nine times.

THE ASSERTS ARE THE PART THAT NEEDED A SECOND PASS. The first revision shipped
four aggregate asserts and an adversarial review broke them by counterexample:
swapping retiredSeq and completedFrameSerial inside LinkProgress compiled
clean, passed all four, and passed the runtime layout test too - because
appliedSeq stays at offset 0. Three wire-visible offsets on a shared page were
pinned by nothing. Nothing else in the tree covers it either: the ABI
fingerprint takes no input from this page, and sizeof(RingControl) is 4096
whatever the members do, because alignas(4096) makes it so.

So the block is now per-field as CONTRACT-P6 §8.3 actually asked, plus a
standard-layout precondition, a tail bound that alignas cannot satisfy for us,
and the LinkEventFlags mirror. Each was run red: the field swap reds the
per-field asserts by name, a fifth member inside LinkProgress reds sizeof.

One of them cannot go red, and Ring.h now says so rather than implying
otherwise. A SMALL member inserted between Progress and serverEpoch lands at
offset 288, inside Progress's own line, and serverEpoch's alignas(64) absorbs
up to 32 bytes of it - so both line asserts still pass. The review's proposed
fix does not catch it either; I tried it. A gate nobody can falsify is worse
than no gate, so the limit is written down beside the asserts.

Also: RingTest.ControlPageLayoutIsTheSharedContract said nothing about the one
field this package moved, and now asserts the producer/consumer separation and
the event flags' exclusion itself rather than borrowing the claim from Ring.h.
Ring.h grew 30 lines, so the Ring.h:NN citations this package and c6 authored
are re-anchored to named blocks instead of line numbers. ARCHITECTURE.md §11.2
is updated for the new layout, and a factual error introduced in the previous
revision of that paragraph - submittedSeq annotated as releasing staging, which
retiredSeq does - is corrected.

Gates: pull 0 added / 0 removed, .text size 10822147 unchanged; split 0/0,
.text size 11681395 unchanged; unit suite 2326/2328 with the same two
pre-existing environmental failures as the baseline. The five edits behind
MGITEST_SPLIT_RUNTIME_PEEK were proven compiled by a red-once, not assumed.

Note on the .text claim: that is a SIZE, not the bytes - field offsets moved,
so immediates moved with them. And this page's layout IS observable across
binaries; what is unobservable is behaviour within one binary. The contract
already books the layout change deliberately, before spawn exists to care.
2026-09-21 09:23:48 -04:00
swung0x48 99649a9087 [MG_Remote] (P6/lk): the pre-flight itemisation, and a correction to c6's regroup design
CONTRACT-P6 §8.5 requires lk to publish its blast radius before it starts,
because an earlier draft's ratchet baseline was wrong by a factor of three and
a ratchet installed on the wrong number is red on its first CI run for reasons
that have nothing to do with a leak. The itemisation is mechanical, per file,
with the grep spelling stated: 250 seam-identifier references outside
Transport/ across 23 files (127 production, 123 test), and 44 member-access
sites that the regroup itself must rewrite, across 9 files - 17 of them in
MG_IntegrationTest, which is why lk builds with the integration tests ON.
The earlier build did not, and a missed edit there would have been green.

THE CORRECTION. Measuring that radius turned up a mistake in c6, landed two
commits ago. §8.3 and ILink.h both said lk should promote eventRingFull and
eventDropped into the watermark group and called the result single-writer.
eventRingFull has TWO writers: the server latches it with store(1) when
SEG_EVENT fills (EventRing.h:140) and the CLIENT clears it with an exchange(0)
on EVERY DRAIN (:200). Promoting it would put a per-drain read-modify-write by
the client on the same cache line as appliedSeq - which the server
release-stores every 64 records and the client's spin predicate reads 1e5-1e6
times a frame. That is new false sharing on the hottest line in the system,
and it would land in exactly the numbers this package's gate has to hold
constant. It is also the same class of error the earlier draft made with
submittedSeq, one field over.

So LinkProgress is now the four consumer-written watermarks and nothing else -
genuinely single-writer - and the two flags stay in the doorbell/generation
group, where mixed writers already live beside consumerParked/producerParked,
crossing the seam through a separate EventFlags() accessor on its own line.
The contract records the correction and its reasoning rather than quietly
restating the rule.

ALSO CORRECTED, from running the baseline suite: a6's link-experiment report
said the silently-empty build stamp had "no log and no gate to say so". That
is false. SessionHandshakeTest.cpp:140 asserts BuildStamp[0] != '\0' and :180
asserts that perturbing it changes the fingerprint, and BOTH GO RED in this
configuration right now - the worktree's .git holds a Windows path WSL's git
cannot resolve, so GIT_COMMIT_HASH_SHORT is "". The finding is therefore
stronger, not weaker: the gate that catches it already exists and is already
red. What is silent is the BUILD, and hs owes a named configure-time failure
rather than a new gate.

Baseline for lk, recorded: 3878 tests, two failures, both environmental and
understood - the fingerprint one above, and PipeCatalogue.FrontendNeverTakes-
AnApplierAddress, which walks up from cwd looking for MobileGL/MG_Impl and so
fails for any out-of-source build directory.

No behaviour change. The headers are still included by nothing.
2026-09-21 08:54:48 -04:00
swung0x48 b6293957e9 [MG_Remote] (P6/c6): the normative contract, and the data-plane seam declared but not wired
c6 is P6's contract package. It lands CONTRACT-P6.md plus two headers and
changes nothing: no caller moves, no byte moves on any wire.

THE CONTRACT supersedes P6-CONTRACT-DRAFT.md, which was written before its own
audit and is wrong in four places - the ABI row, "the latch is implemented
unchanged", "cp is transport only", and "a transport swap and nothing else".
a6 measured all four. It carries 21 rulings, of which the load-bearing one is
D1: the spawned child keeps Transport == Spawn and anti-recursion becomes a
separate Dial axis. ARCHITECTURE.md:488's literal instruction is amended,
because 226 live lines test `Transport != Monolith` to select the SERVER arm
and forcing Monolith flips 148 MG_Backend lines to frontend glue in the one
process that has no frontend.

That ruling can be stated normatively without changing behaviour because of
something a6 found and the contract exploits: TransportMode::Spawn is
UNREACHABLE at this head. ConfigLoader.cpp:344 is the only writer for the
string "spawn" and it writes Monolith. So every rule conditioned on
Transport == Spawn constrains a state no process can enter - and the same fact
is why ConfigLoader's named refusal cannot currently falsify itself, which is
sm's to fix and §9.5's arm-proof gate to catch.

THE SEAM. ITransport says in its own header that the hot path bypasses it;
below it there was no interface of any kind, because Ring.h, ReplySlot.h and
EventRing.h each take a raw void* plus a capacity. ILink.h is that missing
thing. It is setup-time, not hot-path: no method is called per record or inside
a spin predicate, because Doorbell::Wait's predicate runs 1e5-1e6 times a frame
reading watermarks straight off the page, and because the only lane that could
measure a virtual there has four times the dispersion of the effect - a gate
that can only print "within noise" is ID-124 one level down.

StreamLink.h is declared and implemented by nobody, and that is its job: ILink's
adequacy is a claim only a compile can settle. Writing it forced three methods
the brief did not name - ResolveSpan, FlushProgress, and the reply pair - so it
has already earned its keep. Delete one ILink method it declares and the build
goes red.

Deliberately NOT here, each for a reason the contract records: Watermark::Reached
is not redefined (SessionRings.h owns it); Ring.cpp's reserve/pop bodies are not
relocated into a header that would then need the logger; RingControl is not
regrouped, because that rename is 109 production sites with five behind a
non-Android #if, so it is lk's with a published itemisation; and the schema rows
are specified but left to hs, because deprecating SurfaceReply.defaultFb while
appending `result` changes CreateSurfaceReply's arity and the existing
six-argument call would silently bind its defaultFb 0 to result.

INERTNESS IS STRUCTURAL, not measured. Zero object dependency files reference
either header, zero build rules do, and nothing in the tree includes them.
Verified anyway on both flavours: pull and split each have 0 symbols added,
0 removed, and .text byte-identical (10822147 / 11681395). check_doc_citations
resolves all 61 citations.
2026-09-21 08:35:50 -04:00
swung0x48 c25a776090 [Docs] (Disaggregated): P6 a6 close-out, and the end-state review it was re-scoped from
a6 is the read-only audit P6-SPAWN-PLAN puts first. Seven static items, each
produced by one auditor and re-resolved line by line by one adversarial
verifier; 198 rows, 65 of them contested, and only the surviving ones carry
into the report. Item 5 - can the server link without MG_Impl - is the one
experiment, and it is answered.

The headline is that P6's transport half is smaller than the plan assumed
(fc and fs really did land) and its process/role half is much larger. Four
statements in P6-SPAWN-PLAN / P6-CONTRACT-DRAFT are false at this head:

  - `spawn` is not an observable state at all. ConfigLoader.cpp:339-346
    recognises it and then writes Transport = Monolith, so "the child was
    forced to monolith" and "the parent fell back" are the same value and
    that named refusal cannot falsify itself. Meanwhile 226 live lines test
    `Transport != Monolith` to select the SERVER arm, 148 of them in
    MG_Backend - so ARCHITECTURE.md:488's rule, implemented literally, flips
    the whole backend to frontend glue in the one process with no frontend.
  - `cp` is not transport-only: the InitCapabilities REQUEST has no wire form,
    and the codec refuses to encode the kind a live client still calls.
  - the device-lost latch is not implemented; it was booked as carriage.
  - the server cannot link without MG_Impl: 184 symbols, 177 of them real
    function calls, owed 101 to P7 / 14 to P3b-P4b / 60 shared / 6 to P6.
    There is also no module target to link, so the question could not have
    been answered by editing a link line.

Four more blockers the earlier review missed: the apply thread aborts in
ConsumerDoorbell before popping its first record (contract 3.9's ruling, which
lk should retire rather than amend); neither side of the control plane is
wired to a socket; the control-reply wait has no deadline, so a frozen peer
wedges the GL thread inside eglMakeCurrent; and SurfaceReply cannot carry a
MobileGLResult, so a client cannot tell a dead server from a legitimate
refusal.

The abort census makes dl's funnel a redesign rather than a rename: 92 sites,
30 distinct family words against a 7-value wire enum, and three unbounded
peer-driven allocations with no try/catch anywhere under MG_Remote.

P6-ENDSTATE-REVIEW.md is the re-examination of P6 against the gfxstream-shaped
end state that produced a6's three added items. a6 overturned four of its own
claims - the MG_State eight's shape, the abort count, m_windowHandle's
injection points and a predicate name collision - and those are corrected in
place rather than left to be rediscovered.

Docs only. No code, no gate, no behaviour.
2026-09-21 07:49:37 -04:00
1102 changed files with 447539 additions and 24816 deletions
+27 -2
View File
@@ -61,15 +61,40 @@ if ! grep -Eq '^lib/[^/]+/libMobileGL\.so$' <<<"$plugin_contents"; then
exit 1
fi
# The render server is not trace-only: both flavors carry the control screen, the offscreen
# service, the on-screen display activity, and (split native builds) the supervisor executable.
# The plugin APK ships to users, so its build keeps the three components private (manifest
# placeholder serverComponentsExported=false); only the debuggable trace APK exports them for adb.
for component in ServerControlActivity MobileGLServerService MobileGLDisplayActivity; do
require "$component" "$plugin_manifest" "plugin render server component ($component)"
require "$component" "$trace_manifest" "trace render server component ($component)"
done
if [[ "${MOBILEGL_CI_DISAGGREGATED:-OFF}" == "ON" ]]; then
trace_contents=$(unzip -Z1 "$trace_apk")
for entry in "plugin:${plugin_contents}" "trace:${trace_contents}"; do
if ! grep -Eq '^lib/[^/]+/libMobileGLServer\.so$' <<<"${entry#*:}"; then
echo "::error::${entry%%:*} APK does not contain libMobileGLServer.so (both flavors ship the render server)" >&2
exit 1
fi
done
fi
require 'top.mobilegl.plugin.trace' "$trace_manifest" 'trace package name'
require 'top.mobilegl.plugin.TRACE_REPLAY' "$trace_manifest" 'trace replay action'
if grep -Fq 'fclPlugin' <<<"$trace_manifest"; then
echo '::error::Trace APK must not advertise renderer-plugin metadata' >&2
exit 1
fi
if grep -Fq 'android.intent.action.MAIN' <<<"$trace_manifest"; then
echo '::error::Trace APK must not expose a launcher activity' >&2
# The trace APK IS a launchable app now: PostActivity (the POST self-test screen) is its
# entry icon and the bottom navigation bar switches to the render server screen. What is
# left of the old "no launcher" contract is the count - exactly ONE MAIN activity, so a
# screen that accidentally grows a launcher entry (ServerControlActivity was one for a day)
# reds here instead of shipping two icons.
main_count=$(grep -Fc 'android.intent.action.MAIN' <<<"$trace_manifest")
if [[ "$main_count" -ne 1 ]]; then
echo "::error::Trace APK must expose exactly one launcher activity (PostActivity); found ${main_count} android.intent.action.MAIN" >&2
exit 1
fi
require 'PostActivity' "$trace_manifest" 'trace launcher activity (PostActivity)'
echo 'Validated unified MobileGL plugin APK and isolated trace APK.'
+15 -4
View File
@@ -11,8 +11,19 @@ on:
# someone remembering to dispatch the workflow by hand.
- feat/disaggregated
- codex/fix-github-ci
# A docs-only push builds nothing this workflow tests (231 emulator legs per run). The Test
# workflow keeps running for docs: it hosts the doc citation lint.
paths-ignore:
- 'docs/**'
- '**/*.md'
workflow_dispatch:
# ONE LIVE RUN PER BRANCH, as in test.yml: a superseded push is cancelled rather than left queued
# behind its successor; a manual dispatch gets a group of its own; dev queues rather than cancels.
concurrency:
group: ${{ github.workflow }}-${{ github.event_name == 'workflow_dispatch' && github.run_id || github.ref }}
cancel-in-progress: ${{ github.event_name == 'push' && github.ref != 'refs/heads/dev' }}
jobs:
build:
runs-on: ubuntu-latest
@@ -353,7 +364,7 @@ jobs:
--avd-name "${AVD_NAME}"
retrace:
name: retrace (${{ matrix.backend.name }} / ${{ matrix.lane || 'monolith-control' }} / ${{ matrix.transport || 'monolith' }}, ${{ matrix.case.name }})
name: retrace (${{ matrix.backend.name }} / ${{ matrix.transport || 'monolith' }}, ${{ matrix.case.name }})
runs-on: ubuntu-latest
needs:
- build
@@ -374,7 +385,7 @@ jobs:
max-parallel: 4
matrix: ${{ fromJSON(needs.trace-cases.outputs.android) }}
steps:
- name: Set Swap Space
- name: Set swap space
uses: pierotofy/set-swap-space@v1.0
with:
swap-size-gb: 8
@@ -443,7 +454,7 @@ jobs:
--profile pixel_6 \
--avd-name "${AVD_NAME}"
- name: Launch Emulator
- name: Launch emulator
run: |
sh android-plugin/run-avd-ci.sh start \
--avd-name "${AVD_NAME}" \
@@ -600,7 +611,7 @@ jobs:
free -h > android-retrace-result/diagnostics/host-memory.txt 2>&1 || true
sudo dmesg -T 2>/dev/null | tail -300 > android-retrace-result/diagnostics/host-dmesg.txt || true
- name: Stop Emulator
- name: Stop emulator
if: always()
run: |
sh android-plugin/run-avd-ci.sh stop \
File diff suppressed because it is too large Load Diff
+1
View File
@@ -28,3 +28,4 @@ tools/trace_replay/work/
__pycache__/
*.py[cod]
/.gradle
/.trace-work/
+538
View File
@@ -0,0 +1,538 @@
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
* HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
* WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*
* Authors:
* Benjamin Franzke <benjaminfranzke@googlemail.com>
*/
#ifndef _GBM_H_
#define _GBM_H_
#define __GBM__ 1
#include <stddef.h>
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
/**
* \file gbm.h
* \brief Generic Buffer Manager
*/
struct gbm_device;
struct gbm_bo;
struct gbm_surface;
/**
* \mainpage The Generic Buffer Manager
*
* This module provides an abstraction that the caller can use to request a
* buffer from the underlying memory management system for the platform.
*
* This allows the creation of portable code whilst still allowing access to
* the underlying memory manager.
*/
/**
* Abstraction representing the handle to a buffer allocated by the
* manager
*/
union gbm_bo_handle {
void *ptr;
int32_t s32;
uint32_t u32;
int64_t s64;
uint64_t u64;
};
/** Format of the allocated buffer */
enum gbm_bo_format {
/** RGB with 8 bits per channel in a 32 bit value */
GBM_BO_FORMAT_XRGB8888,
/** ARGB with 8 bits per channel in a 32 bit value */
GBM_BO_FORMAT_ARGB8888
};
/**
* The FourCC format codes are taken from the drm_fourcc.h definition, and
* re-namespaced. New GBM formats must not be added, unless they are
* identical ports from drm_fourcc.
*/
#define __gbm_fourcc_code(a,b,c,d) ((uint32_t)(a) | ((uint32_t)(b) << 8) | \
((uint32_t)(c) << 16) | ((uint32_t)(d) << 24))
#define GBM_FORMAT_BIG_ENDIAN (1<<31) /* format is big endian instead of little endian */
/* color index */
#define GBM_FORMAT_C8 __gbm_fourcc_code('C', '8', ' ', ' ') /* [7:0] C */
/* 8 bpp Red */
#define GBM_FORMAT_R8 __gbm_fourcc_code('R', '8', ' ', ' ') /* [7:0] R */
/* 16 bpp Red */
#define GBM_FORMAT_R16 __gbm_fourcc_code('R', '1', '6', ' ') /* [15:0] R little endian */
/* 16 bpp RG */
#define GBM_FORMAT_GR88 __gbm_fourcc_code('G', 'R', '8', '8') /* [15:0] G:R 8:8 little endian */
/* 32 bpp RG */
#define GBM_FORMAT_RG1616 __gbm_fourcc_code('R', 'G', '3', '2') /* [31:0] R:G 16:16 little endian */
#define GBM_FORMAT_GR1616 __gbm_fourcc_code('G', 'R', '3', '2') /* [31:0] G:R 16:16 little endian */
/* 8 bpp RGB */
#define GBM_FORMAT_RGB332 __gbm_fourcc_code('R', 'G', 'B', '8') /* [7:0] R:G:B 3:3:2 */
#define GBM_FORMAT_BGR233 __gbm_fourcc_code('B', 'G', 'R', '8') /* [7:0] B:G:R 2:3:3 */
/* 16 bpp RGB */
#define GBM_FORMAT_XRGB4444 __gbm_fourcc_code('X', 'R', '1', '2') /* [15:0] x:R:G:B 4:4:4:4 little endian */
#define GBM_FORMAT_XBGR4444 __gbm_fourcc_code('X', 'B', '1', '2') /* [15:0] x:B:G:R 4:4:4:4 little endian */
#define GBM_FORMAT_RGBX4444 __gbm_fourcc_code('R', 'X', '1', '2') /* [15:0] R:G:B:x 4:4:4:4 little endian */
#define GBM_FORMAT_BGRX4444 __gbm_fourcc_code('B', 'X', '1', '2') /* [15:0] B:G:R:x 4:4:4:4 little endian */
#define GBM_FORMAT_ARGB4444 __gbm_fourcc_code('A', 'R', '1', '2') /* [15:0] A:R:G:B 4:4:4:4 little endian */
#define GBM_FORMAT_ABGR4444 __gbm_fourcc_code('A', 'B', '1', '2') /* [15:0] A:B:G:R 4:4:4:4 little endian */
#define GBM_FORMAT_RGBA4444 __gbm_fourcc_code('R', 'A', '1', '2') /* [15:0] R:G:B:A 4:4:4:4 little endian */
#define GBM_FORMAT_BGRA4444 __gbm_fourcc_code('B', 'A', '1', '2') /* [15:0] B:G:R:A 4:4:4:4 little endian */
#define GBM_FORMAT_XRGB1555 __gbm_fourcc_code('X', 'R', '1', '5') /* [15:0] x:R:G:B 1:5:5:5 little endian */
#define GBM_FORMAT_XBGR1555 __gbm_fourcc_code('X', 'B', '1', '5') /* [15:0] x:B:G:R 1:5:5:5 little endian */
#define GBM_FORMAT_RGBX5551 __gbm_fourcc_code('R', 'X', '1', '5') /* [15:0] R:G:B:x 5:5:5:1 little endian */
#define GBM_FORMAT_BGRX5551 __gbm_fourcc_code('B', 'X', '1', '5') /* [15:0] B:G:R:x 5:5:5:1 little endian */
#define GBM_FORMAT_ARGB1555 __gbm_fourcc_code('A', 'R', '1', '5') /* [15:0] A:R:G:B 1:5:5:5 little endian */
#define GBM_FORMAT_ABGR1555 __gbm_fourcc_code('A', 'B', '1', '5') /* [15:0] A:B:G:R 1:5:5:5 little endian */
#define GBM_FORMAT_RGBA5551 __gbm_fourcc_code('R', 'A', '1', '5') /* [15:0] R:G:B:A 5:5:5:1 little endian */
#define GBM_FORMAT_BGRA5551 __gbm_fourcc_code('B', 'A', '1', '5') /* [15:0] B:G:R:A 5:5:5:1 little endian */
#define GBM_FORMAT_RGB565 __gbm_fourcc_code('R', 'G', '1', '6') /* [15:0] R:G:B 5:6:5 little endian */
#define GBM_FORMAT_BGR565 __gbm_fourcc_code('B', 'G', '1', '6') /* [15:0] B:G:R 5:6:5 little endian */
/* 24 bpp RGB */
#define GBM_FORMAT_RGB888 __gbm_fourcc_code('R', 'G', '2', '4') /* [23:0] R:G:B little endian */
#define GBM_FORMAT_BGR888 __gbm_fourcc_code('B', 'G', '2', '4') /* [23:0] B:G:R little endian */
/* 32 bpp RGB */
#define GBM_FORMAT_XRGB8888 __gbm_fourcc_code('X', 'R', '2', '4') /* [31:0] x:R:G:B 8:8:8:8 little endian */
#define GBM_FORMAT_XBGR8888 __gbm_fourcc_code('X', 'B', '2', '4') /* [31:0] x:B:G:R 8:8:8:8 little endian */
#define GBM_FORMAT_RGBX8888 __gbm_fourcc_code('R', 'X', '2', '4') /* [31:0] R:G:B:x 8:8:8:8 little endian */
#define GBM_FORMAT_BGRX8888 __gbm_fourcc_code('B', 'X', '2', '4') /* [31:0] B:G:R:x 8:8:8:8 little endian */
#define GBM_FORMAT_ARGB8888 __gbm_fourcc_code('A', 'R', '2', '4') /* [31:0] A:R:G:B 8:8:8:8 little endian */
#define GBM_FORMAT_ABGR8888 __gbm_fourcc_code('A', 'B', '2', '4') /* [31:0] A:B:G:R 8:8:8:8 little endian */
#define GBM_FORMAT_RGBA8888 __gbm_fourcc_code('R', 'A', '2', '4') /* [31:0] R:G:B:A 8:8:8:8 little endian */
#define GBM_FORMAT_BGRA8888 __gbm_fourcc_code('B', 'A', '2', '4') /* [31:0] B:G:R:A 8:8:8:8 little endian */
#define GBM_FORMAT_XRGB2101010 __gbm_fourcc_code('X', 'R', '3', '0') /* [31:0] x:R:G:B 2:10:10:10 little endian */
#define GBM_FORMAT_XBGR2101010 __gbm_fourcc_code('X', 'B', '3', '0') /* [31:0] x:B:G:R 2:10:10:10 little endian */
#define GBM_FORMAT_RGBX1010102 __gbm_fourcc_code('R', 'X', '3', '0') /* [31:0] R:G:B:x 10:10:10:2 little endian */
#define GBM_FORMAT_BGRX1010102 __gbm_fourcc_code('B', 'X', '3', '0') /* [31:0] B:G:R:x 10:10:10:2 little endian */
#define GBM_FORMAT_ARGB2101010 __gbm_fourcc_code('A', 'R', '3', '0') /* [31:0] A:R:G:B 2:10:10:10 little endian */
#define GBM_FORMAT_ABGR2101010 __gbm_fourcc_code('A', 'B', '3', '0') /* [31:0] A:B:G:R 2:10:10:10 little endian */
#define GBM_FORMAT_RGBA1010102 __gbm_fourcc_code('R', 'A', '3', '0') /* [31:0] R:G:B:A 10:10:10:2 little endian */
#define GBM_FORMAT_BGRA1010102 __gbm_fourcc_code('B', 'A', '3', '0') /* [31:0] B:G:R:A 10:10:10:2 little endian */
/* 64 bpp RGB */
#define GBM_FORMAT_XBGR16161616 __gbm_fourcc_code('X', 'B', '4', '8') /* [63:0] x:B:G:R 16:16:16:16 little endian */
#define GBM_FORMAT_ABGR16161616 __gbm_fourcc_code('A', 'B', '4', '8') /* [63:0] A:B:G:R 16:16:16:16 little endian */
/*
* Floating point 64bpp RGB
* IEEE 754-2008 binary16 half-precision float
* [15:0] sign:exponent:mantissa 1:5:10
*/
#define GBM_FORMAT_XBGR16161616F __gbm_fourcc_code('X', 'B', '4', 'H') /* [63:0] x:B:G:R 16:16:16:16 little endian */
#define GBM_FORMAT_ABGR16161616F __gbm_fourcc_code('A', 'B', '4', 'H') /* [63:0] A:B:G:R 16:16:16:16 little endian */
/* packed YCbCr */
#define GBM_FORMAT_YUYV __gbm_fourcc_code('Y', 'U', 'Y', 'V') /* [31:0] Cr0:Y1:Cb0:Y0 8:8:8:8 little endian */
#define GBM_FORMAT_YVYU __gbm_fourcc_code('Y', 'V', 'Y', 'U') /* [31:0] Cb0:Y1:Cr0:Y0 8:8:8:8 little endian */
#define GBM_FORMAT_UYVY __gbm_fourcc_code('U', 'Y', 'V', 'Y') /* [31:0] Y1:Cr0:Y0:Cb0 8:8:8:8 little endian */
#define GBM_FORMAT_VYUY __gbm_fourcc_code('V', 'Y', 'U', 'Y') /* [31:0] Y1:Cb0:Y0:Cr0 8:8:8:8 little endian */
#define GBM_FORMAT_AYUV __gbm_fourcc_code('A', 'Y', 'U', 'V') /* [31:0] A:Y:Cb:Cr 8:8:8:8 little endian */
/*
* 2 plane YCbCr
* index 0 = Y plane, [7:0] Y
* index 1 = Cr:Cb plane, [15:0] Cr:Cb little endian
* or
* index 1 = Cb:Cr plane, [15:0] Cb:Cr little endian
*/
#define GBM_FORMAT_NV12 __gbm_fourcc_code('N', 'V', '1', '2') /* 2x2 subsampled Cr:Cb plane */
#define GBM_FORMAT_NV21 __gbm_fourcc_code('N', 'V', '2', '1') /* 2x2 subsampled Cb:Cr plane */
#define GBM_FORMAT_NV16 __gbm_fourcc_code('N', 'V', '1', '6') /* 2x1 subsampled Cr:Cb plane */
#define GBM_FORMAT_NV61 __gbm_fourcc_code('N', 'V', '6', '1') /* 2x1 subsampled Cb:Cr plane */
/*
* 3 plane YCbCr
* index 0: Y plane, [7:0] Y
* index 1: Cb plane, [7:0] Cb
* index 2: Cr plane, [7:0] Cr
* or
* index 1: Cr plane, [7:0] Cr
* index 2: Cb plane, [7:0] Cb
*/
#define GBM_FORMAT_YUV410 __gbm_fourcc_code('Y', 'U', 'V', '9') /* 4x4 subsampled Cb (1) and Cr (2) planes */
#define GBM_FORMAT_YVU410 __gbm_fourcc_code('Y', 'V', 'U', '9') /* 4x4 subsampled Cr (1) and Cb (2) planes */
#define GBM_FORMAT_YUV411 __gbm_fourcc_code('Y', 'U', '1', '1') /* 4x1 subsampled Cb (1) and Cr (2) planes */
#define GBM_FORMAT_YVU411 __gbm_fourcc_code('Y', 'V', '1', '1') /* 4x1 subsampled Cr (1) and Cb (2) planes */
#define GBM_FORMAT_YUV420 __gbm_fourcc_code('Y', 'U', '1', '2') /* 2x2 subsampled Cb (1) and Cr (2) planes */
#define GBM_FORMAT_YVU420 __gbm_fourcc_code('Y', 'V', '1', '2') /* 2x2 subsampled Cr (1) and Cb (2) planes */
#define GBM_FORMAT_YUV422 __gbm_fourcc_code('Y', 'U', '1', '6') /* 2x1 subsampled Cb (1) and Cr (2) planes */
#define GBM_FORMAT_YVU422 __gbm_fourcc_code('Y', 'V', '1', '6') /* 2x1 subsampled Cr (1) and Cb (2) planes */
#define GBM_FORMAT_YUV444 __gbm_fourcc_code('Y', 'U', '2', '4') /* non-subsampled Cb (1) and Cr (2) planes */
#define GBM_FORMAT_YVU444 __gbm_fourcc_code('Y', 'V', '2', '4') /* non-subsampled Cr (1) and Cb (2) planes */
struct gbm_format_name_desc {
char name[5];
};
/**
* Flags to indicate the intended use for the buffer - these are passed into
* gbm_bo_create(). The caller must set the union of all the flags that are
* appropriate
*
* \sa Use gbm_device_is_format_supported() to check if the combination of format
* and use flags are supported
*/
enum gbm_bo_flags {
/**
* Buffer is going to be presented to the screen using an API such as KMS
*/
GBM_BO_USE_SCANOUT = (1 << 0),
/**
* Buffer is going to be used as cursor
*/
GBM_BO_USE_CURSOR = (1 << 1),
/**
* Deprecated
*/
GBM_BO_USE_CURSOR_64X64 = GBM_BO_USE_CURSOR,
/**
* Buffer is to be used for rendering - for example it is going to be used
* as the storage for a color buffer
*/
GBM_BO_USE_RENDERING = (1 << 2),
/**
* Buffer can be used for gbm_bo_write. This is guaranteed to work
* with GBM_BO_USE_CURSOR, but may not work for other combinations.
*/
GBM_BO_USE_WRITE = (1 << 3),
/**
* Buffer is linear, i.e. not tiled.
*/
GBM_BO_USE_LINEAR = (1 << 4),
/**
* Buffer is protected, i.e. encrypted and not readable by CPU or any
* other non-secure / non-trusted components nor by non-trusted OpenGL,
* OpenCL, and Vulkan applications.
*/
GBM_BO_USE_PROTECTED = (1 << 5),
/**
* The buffer will be used for front buffer rendering. On some
* platforms this may (for example) disable framebuffer compression
* to avoid problems with compression flags data being out of sync
* with pixel data.
*/
GBM_BO_USE_FRONT_RENDERING = (1 << 6),
/**
* Allow the driver to select fixed-rate compression parameters.
*/
GBM_BO_FIXED_COMPRESSION_DEFAULT = (1 << 7),
/**
* Fixed-rate compression: at least 1bpc, less than 2bpc
*/
GBM_BO_FIXED_COMPRESSION_1BPC = (2 << 7),
/**
* Fixed-rate compression: at least 2bpc, less than 3bpc
*/
GBM_BO_FIXED_COMPRESSION_2BPC = (3 << 7),
/**
* Fixed-rate compression: at least 3bpc, less than 4bpc
*/
GBM_BO_FIXED_COMPRESSION_3BPC = (4 << 7),
/**
* Fixed-rate compression: at least 4bpc, less than 5bpc
*/
GBM_BO_FIXED_COMPRESSION_4BPC = (5 << 7),
/**
* Fixed-rate compression: at least 5bpc, less than 6bpc
*/
GBM_BO_FIXED_COMPRESSION_5BPC = (6 << 7),
/**
* Fixed-rate compression: at least 6bpc, less than 7bpc
*/
GBM_BO_FIXED_COMPRESSION_6BPC = (7 << 7),
/**
* Fixed-rate compression: at least 7bpc, less than 8bpc
*/
GBM_BO_FIXED_COMPRESSION_7BPC = (8 << 7),
/**
* Fixed-rate compression: at least 8bpc, less than 9bpc
*/
GBM_BO_FIXED_COMPRESSION_8BPC = (9 << 7),
/**
* Fixed-rate compression: at least 9bpc, less than 10bpc
*/
GBM_BO_FIXED_COMPRESSION_9BPC = (10 << 7),
/**
* Fixed-rate compression: at least 10bpc, less than 11bpc
*/
GBM_BO_FIXED_COMPRESSION_10BPC = (11 << 7),
/**
* Fixed-rate compression: at least 11bpc, less than 12bpc
*/
GBM_BO_FIXED_COMPRESSION_11BPC = (12 << 7),
/**
* Fixed-rate compression: at least 12bpc, no maximum rate
*/
GBM_BO_FIXED_COMPRESSION_12BPC = (13 << 7),
/* next available value is (1 << 11) */
};
#define GBM_BO_FIXED_COMPRESSION_MASK (((1 << 11) - 1) & ~((1 << 7) - 1))
int
gbm_device_get_fd(struct gbm_device *gbm);
const char *
gbm_device_get_backend_name(struct gbm_device *gbm);
int
gbm_device_is_format_supported(struct gbm_device *gbm,
uint32_t format, uint32_t flags);
int
gbm_device_get_format_modifier_plane_count(struct gbm_device *gbm,
uint32_t format,
uint64_t modifier);
void
gbm_device_destroy(struct gbm_device *gbm);
struct gbm_device *
gbm_create_device(int fd);
struct gbm_bo *
gbm_bo_create(struct gbm_device *gbm,
uint32_t width, uint32_t height,
uint32_t format, uint32_t flags);
struct gbm_bo *
gbm_bo_create_with_modifiers(struct gbm_device *gbm,
uint32_t width, uint32_t height,
uint32_t format,
const uint64_t *modifiers,
const unsigned int count);
struct gbm_bo *
gbm_bo_create_with_modifiers2(struct gbm_device *gbm,
uint32_t width, uint32_t height,
uint32_t format,
const uint64_t *modifiers,
const unsigned int count,
uint32_t flags);
#define GBM_BO_IMPORT_WL_BUFFER 0x5501
#define GBM_BO_IMPORT_EGL_IMAGE 0x5502
#define GBM_BO_IMPORT_FD 0x5503
#define GBM_BO_IMPORT_FD_MODIFIER 0x5504
struct gbm_import_fd_data {
int fd;
uint32_t width;
uint32_t height;
uint32_t stride;
uint32_t format;
};
#define GBM_MAX_PLANES 4
struct gbm_import_fd_modifier_data {
uint32_t width;
uint32_t height;
uint32_t format;
uint32_t num_fds;
int fds[GBM_MAX_PLANES];
int strides[GBM_MAX_PLANES];
int offsets[GBM_MAX_PLANES];
uint64_t modifier;
};
struct gbm_bo *
gbm_bo_import(struct gbm_device *gbm, uint32_t type,
void *buffer, uint32_t flags);
/**
* Flags to indicate the type of mapping for the buffer - these are
* passed into gbm_bo_map(). The caller must set the union of all the
* flags that are appropriate.
*
* These flags are independent of the GBM_BO_USE_* creation flags. However,
* mapping the buffer may require copying to/from a staging buffer.
*
* See also: pipe_map_flags
*/
enum gbm_bo_transfer_flags {
/**
* Buffer contents read back (or accessed directly) at transfer
* create time.
*/
GBM_BO_TRANSFER_READ = (1 << 0),
/**
* Buffer contents will be written back at unmap time
* (or modified as a result of being accessed directly).
*/
GBM_BO_TRANSFER_WRITE = (1 << 1),
/**
* Read/modify/write
*/
GBM_BO_TRANSFER_READ_WRITE = (GBM_BO_TRANSFER_READ | GBM_BO_TRANSFER_WRITE),
};
void *
gbm_bo_map(struct gbm_bo *bo,
uint32_t x, uint32_t y, uint32_t width, uint32_t height,
uint32_t flags, uint32_t *stride, void **map_data);
void
gbm_bo_unmap(struct gbm_bo *bo, void *map_data);
uint32_t
gbm_bo_get_width(struct gbm_bo *bo);
uint32_t
gbm_bo_get_height(struct gbm_bo *bo);
uint32_t
gbm_bo_get_stride(struct gbm_bo *bo);
uint32_t
gbm_bo_get_stride_for_plane(struct gbm_bo *bo, int plane);
uint32_t
gbm_bo_get_format(struct gbm_bo *bo);
uint32_t
gbm_bo_get_bpp(struct gbm_bo *bo);
uint32_t
gbm_bo_get_offset(struct gbm_bo *bo, int plane);
struct gbm_device *
gbm_bo_get_device(struct gbm_bo *bo);
union gbm_bo_handle
gbm_bo_get_handle(struct gbm_bo *bo);
int
gbm_bo_get_fd(struct gbm_bo *bo);
uint64_t
gbm_bo_get_modifier(struct gbm_bo *bo);
int
gbm_bo_get_plane_count(struct gbm_bo *bo);
union gbm_bo_handle
gbm_bo_get_handle_for_plane(struct gbm_bo *bo, int plane);
int
gbm_bo_get_fd_for_plane(struct gbm_bo *bo, int plane);
int
gbm_bo_write(struct gbm_bo *bo, const void *buf, size_t count);
void
gbm_bo_set_user_data(struct gbm_bo *bo, void *data,
void (*destroy_user_data)(struct gbm_bo *, void *));
void *
gbm_bo_get_user_data(struct gbm_bo *bo);
void
gbm_bo_destroy(struct gbm_bo *bo);
struct gbm_surface *
gbm_surface_create(struct gbm_device *gbm,
uint32_t width, uint32_t height,
uint32_t format, uint32_t flags);
struct gbm_surface *
gbm_surface_create_with_modifiers(struct gbm_device *gbm,
uint32_t width, uint32_t height,
uint32_t format,
const uint64_t *modifiers,
const unsigned int count);
struct gbm_surface *
gbm_surface_create_with_modifiers2(struct gbm_device *gbm,
uint32_t width, uint32_t height,
uint32_t format,
const uint64_t *modifiers,
const unsigned int count,
uint32_t flags);
struct gbm_bo *
gbm_surface_lock_front_buffer(struct gbm_surface *surface);
void
gbm_surface_release_buffer(struct gbm_surface *surface, struct gbm_bo *bo);
int
gbm_surface_has_free_buffers(struct gbm_surface *surface);
void
gbm_surface_destroy(struct gbm_surface *surface);
char *
gbm_format_get_name(uint32_t gbm_format, struct gbm_format_name_desc *desc);
#ifdef __cplusplus
}
#endif
#endif
+11
View File
@@ -0,0 +1,11 @@
# glvnd vendor ABI headers
'libeglabi.h' and 'GLdispatchABI.h' from libglvnd (MIT), copied verbatim with their own licence
headers intact.
They are here because MobileGL implements the *vendor* half of glvnd's EGL ABI - the side a library
exposes so the system's 'libEGL.so.1', which IS glvnd, can dispatch to it. The distribution ships
the runtime but not these headers in the base image, and a vendored copy of two MIT headers is a
smaller dependency than a build-time download of a source tree.
Upstream ABI version implemented: 'EGL_VENDOR_ABI_MAJOR_VERSION' 0 (the header pins it).
+141
View File
@@ -0,0 +1,141 @@
/*
* Copyright (c) 2013, NVIDIA CORPORATION.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and/or associated documentation files (the
* "Materials"), to deal in the Materials without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Materials, and to
* permit persons to whom the Materials are furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be included
* unaltered in all copies or substantial portions of the Materials.
* Any additions, deletions, or changes to the original source files
* must be clearly indicated in accompanying documentation.
*
* THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
*/
#include <GL/gl.h>
#if !defined(__GL_DISPATCH_ABI_H)
#define __GL_DISPATCH_ABI_H
#if defined(__cplusplus)
extern "C" {
#endif
/*!
* \defgroup gldispatchabi GL dispatching ABI
*
* This is not a complete ABI, but rather a fragment common to the libEGL and
* libGLX ABIs. Changes to this file should be accompanied by a version bump to
* these client ABIs.
*/
/*!
* Thread-local implementation used by libglvnd. This is passed into the patch
* function callback via the type parameter.
*
* For most architectures, the vendor library can ignore this parameter, since
* it will always be the same value. It's used for systems like ARM, where the
* stubs might be use the ARM or Thumb instruction sets.
*
* The stub type does not make any distinction between TLS and TSD stubs. The
* entire purpose of entrypoint rewriting is to skip the dispatch table in
* libGLdispatch.so, so it doesn't matter how that dispatch table is stored.
*/
enum {
/*!
* Indicates that the stubs aren't defined in assembly. For example, if the
* dispatch stubs are written in C. Vendor libraries generally won't see
* this value.
*/
__GLDISPATCH_STUB_UNKNOWN,
/*!
* Used for stubs on x86 systems.
*/
__GLDISPATCH_STUB_X86,
/*!
* Used for stubs on x86-64 systems.
*/
__GLDISPATCH_STUB_X86_64,
/*!
* Used for stubs on ARMv7, using the Thumb instruction set.
*/
__GLDISPATCH_STUB_ARMV7_THUMB,
/*!
* Used for stubs on ARMv7, using the normal ARM instruction set.
*/
__GLDISPATCH_STUB_ARMV7_ARM,
/*!
* Used for stubs on ARMv8/aarch64.
*/
__GLDISPATCH_STUB_AARCH64,
/*!
* Used for stubs on x32 builds (x86-64 with 32-bit pointers).
*/
__GLDISPATCH_STUB_X32,
/*!
* Used for stubs on PPC64 systems.
*/
__GLDISPATCH_STUB_PPC64,
/*!
* Used for stubs on PPC64LE systems. Same as PPC64, for compatibility.
*/
__GLDISPATCH_STUB_PPC64LE = __GLDISPATCH_STUB_PPC64,
/*!
* Used for stubs on LOONGARCH/loongarch64.
*/
__GLDISPATCH_STUB_LOONGARCH64,
};
/*!
* A callback function called by the vendor library to fetch the address of an
* entrypoint.
*
* The function returns two pointers, one writable and one executable. The two
* pointers may or may not be the same virtual address, but they will both be
* mappings of the same physical memory.
*
* The vendor library should write its entrypoint to the address returned by
* \p writePtr, but should use the address from \p execPtr for things like
* calculating PC-relative offsets.
*
* Note that if this function fails, then the vendor library can still try to
* patch other entrypoints.
*
* Note that on ARM, the low-order bit of both \c execPtr and \p writePtr will
* be zero, even if the stub uses the thumb instruction set. The vendor library
* should use the \c type parameter of \c initiatePatch to determine which
* instruction set to use.
*
* \param funcName The function name.
* \param[out] writePtr The pointer that the vendor library can write to.
* \param[out] execPtr The pointer to the executable code.
* \return GL_TRUE if the entrypoint exists, or GL_FALSE if it doesn't.
*/
typedef GLboolean (*DispatchPatchLookupStubOffset)(const char *funcName,
void **writePtr, const void **execPtr);
#if defined(__cplusplus)
}
#endif
#endif // __GL_DISPATCH_ABI_H
+464
View File
@@ -0,0 +1,464 @@
/*
* Copyright (c) 2013, NVIDIA CORPORATION.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and/or associated documentation files (the
* "Materials"), to deal in the Materials without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Materials, and to
* permit persons to whom the Materials are furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be included
* unaltered in all copies or substantial portions of the Materials.
* Any additions, deletions, or changes to the original source files
* must be clearly indicated in accompanying documentation.
*
* THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
*/
#if !defined(__LIB_EGL_ABI_H)
#define __LIB_EGL_ABI_H
#include <stdint.h>
#include <EGL/egl.h>
#include <EGL/eglext.h>
#include "glvnd/GLdispatchABI.h"
#if defined(__cplusplus)
extern "C" {
#endif
/*!
* \defgroup eglvendorabi EGL Vendor ABI
*
* Definition of ABI exported by libEGL.so to libEGL_VENDOR.so libraries.
*
* Each vendor is associated with three distinct dispatch table types:
*
* - static EGL dispatch table: this is the fixed list of EGL 1.5 entrypoints
* provided by the vendor at load time during the initial handshake.
* - dynamic EGL dispatch table: this is a structure allocated by the API
* library at runtime used to manage EGL extension functions which are not
* present in the static table.
* - core GL dispatch table: this is a structure maintained by the API library
* which contains both GL core (static) and GL extension (dynamic) functions.
*
* @{
*/
/*
* Rendering API handling:
*
* libEGL only supports OpenGL and OpenGL ES, not OpenVG. If OpenVG or any
* other API is added, then the major version number will be incremented.
*
* When the application calls eglBindAPI, libEGL will forward the call to every
* vendor library. In addition, a vendor library can query the current API from
* libEGL using the getCurrentApi callback.
*
* Vendor libraries are not required to support both GL and GLES, but they must
* be able to deal with either one as the current rendering API. If a vendor
* doesn't support the current API, then it should return an error from
* eglCreateContext.
*/
/*!
* Current version of the ABI.
*
* This version number contains a major number in the high-order 16 bits, and
* a minor version number in the low-order 16 bits.
*
* The major version number is incremented when an interface change will break
* backwards compatibility with existing vendor libraries. The minor version
* number is incremented when there's a change but existing vendor libraries
* will still work.
*/
#define EGL_VENDOR_ABI_MAJOR_VERSION ((uint32_t) 0)
#define EGL_VENDOR_ABI_MINOR_VERSION ((uint32_t) 2)
#define EGL_VENDOR_ABI_VERSION ((EGL_VENDOR_ABI_MAJOR_VERSION << 16) | EGL_VENDOR_ABI_MINOR_VERSION)
static inline uint32_t EGL_VENDOR_ABI_GET_MAJOR_VERSION(uint32_t version)
{
return version >> 16;
}
static inline uint32_t EGL_VENDOR_ABI_GET_MINOR_VERSION(uint32_t version)
{
return version & 0xFFFF;
}
/*!
* This opaque structure stores function pointers for EGL extension functions.
* It is allocated at runtime by the API library. Vendor-provided dispatch
* functions retrieve and operate on this structure using the API below.
*/
typedef struct __EGLvendorInfoRec __EGLvendorInfo;
/****************************************************************************
* API library exports *
****************************************************************************/
typedef struct __EGLapiExportsRec {
/************************************************************************
* The following routines are used by vendor-implemented EGL dispatch
* functions to lookup and call into the right vendor.
************************************************************************/
/*!
* This function must be called at the start of every EGL dispatch stub. It
* performs any necessary per-call bookkeeping.
*/
void (* threadInit) (void);
EGLenum (* getCurrentApi)(void);
/*!
* This function retrieves the appropriate current dynamic dispatch table,
* if a GL context is current. Otherwise, this returns NULL.
*/
__EGLvendorInfo *(*getCurrentVendor)(void);
/*!
* This retrieves the current context for this thread.
*/
EGLContext (*getCurrentContext)(void);
/*!
* Returns the current display for this thread.
*/
EGLDisplay (*getCurrentDisplay)(void);
/*!
* Returns the current drawable for this thread.
*/
EGLSurface (*getCurrentSurface)(EGLint readDraw);
/*!
* This function retrieves an entry point from the dynamic dispatch table
* given an index into the table.
*/
__eglMustCastToProperFunctionPointerType (*fetchDispatchEntry) (__EGLvendorInfo *dynDispatch, int index);
/*!
* Sets the last error for the current thread. The error will be returned
* the next time the app calls eglGetError().
*
* This function will override a call to \c setLastVendor, and vice-versa.
*/
void (* setEGLError) (EGLint errorCode);
/*!
* Notifies libEGL about the vendor library that an EGL function is
* dispatched to. This is used to look up the last EGL error code from a
* vendor.
*
* A vendor library may generate an EGL error asynchronously, if it
* offloads some EGL functions onto a worker thread. In that case, calling
* \c setEGLError to set the last error would not work.
*
* When an EGL dispatch stubs finds the vendor library, it must call this
* function before it dispatches the function call.
*
* This function will override a call to \c setEGLError, and vice-versa.
*
* \param vendor The vendor library of the current EGL function call.
* \return EGL_TRUE on success, or EGL_FALSE on failure.
*/
EGLBoolean (* setLastVendor) (__EGLvendorInfo *vendor);
/************************************************************************
* These routines are used by vendor dispatch functions to look up
* and add mappings between various objects and vendors.
************************************************************************/
// As far as I can tell, all EGL functions that take a context or drawable
// also take an EGLDisplay. So, we don't need any extra functions for
// mapping a context/surface/whatever to a EGLDisplay or vendor.
/*!
* Returns the EGL vendor for an EGLDisplay handle.
*/
__EGLvendorInfo *(*getVendorFromDisplay)(EGLDisplay dpy);
/*!
* Returns the EGL vendor for an EGLDeviceEXT handle.
*/
__EGLvendorInfo *(*getVendorFromDevice)(EGLDeviceEXT dev);
/**
* Sets the EGL vendor for an EGLDeviceEXT handle. The dispatch stub for
* any function that returns an EGLDeviceEXT handle should call this.
*
* Supported since ABI version 1.2.
*/
EGLBoolean (* setVendorForDevice)(EGLDeviceEXT dev, __EGLvendorInfo *vendor);
} __EGLapiExports;
/*****************************************************************************
* API library imports *
*****************************************************************************/
/*!
* The enum values accepted by \c __EGLapiImports::getVendorString.
*/
enum
{
/*!
* The set of platform extensions that the vendor supports.
*
* Platform extensions typically just add enums to the existing
* eglGetPlatformDisplay function, so we don't need any additional logic
* or special handling from libEGL itself.
*
* As a result, vendor libraries can add these extensions without having
* to modify libEGL to match.
*/
__EGL_VENDOR_STRING_PLATFORM_EXTENSIONS,
};
/*!
* This structure stores required and optional vendor library callbacks.
*/
typedef struct __EGLapiImportsRec {
/**
* Creates an EGLDisplay. This function is used to handle both
* \c eglGetDisplay and \c eglGetPlatformDisplay.
*
* For some platform types, libEGL may be able to identify a specific
* vendor to use.
*
* If libEGL can't figure out which vendor to use on its own, then it will
* go through the list of available vendor libraries and call this function
* until one succeeds.
*
* If the application calls eglGetPlatformDisplay, then the parameters are
* passed through from the application.
*
* If the application calls eglGetDisplay, and the native display handle is
* not \c EGL_DEFAULT_DISPLAY, then libEGL will try to guess a platform
* type, and will pass that platform to the vendor library.
*
* Lastly, if eglGetDisplay is called with \c EGL_DEFAULT_DISPLAY, then
* libEGL will call into each vendor library with \p platform set to
* \c EGL_NONE. The vendor library can then select a default display to
* return.
*
* In all cases, if the vendor library can't find a matching EGLDisplay,
* then it should return \c EGL_NO_DISPLAY. Any errors should be reported
* through the vendor's \c eglGetError function.
*/
EGLDisplay (* getPlatformDisplay) (EGLenum platform, void *nativeDisplay,
const EGLAttrib *attrib_list);
/*!
* Checks if the vendor library supports a given client API (that is, the
* API value passed to \c eglBindAPI).
*/
EGLBoolean (* getSupportsAPI) (EGLenum api);
/*!
* (OPTIONAL) Returns a string from the vendor library. This is used for
* anything that isn't available from eglQueryString.
*/
const char * (* getVendorString) (int name);
/*
* NOTE: libEGL will load all of the vendor libraries up front, so it
* handles extension functions slightly differently than EGL.
*/
/*!
* This retrieves the pointer to the real EGL or core GL function.
*
* \param procName The name of the function.
* \return A pointer to a function, or \c NULL if the vendor does not
* support the function.
*/
void * (* getProcAddress) (const char *procName);
/*!
* This retrieves vendor-neutral functions which use the
* __EGLdispatchTableDynamic API above to dispatch to the correct vendor.
*
* A vendor library must provide a dispatch function for all EGL display
* extension functions that it supports.
*
* Client extension functions cannot be dispatched based on an EGDisplay,
* so they must be handled in libEGL itself.
*
* \param procName The name of the function.
* \return A pointer to a function, or \c NULL if the vendor does not
* support the function or \p procName is not a EGL display function.
*/
void * (*getDispatchAddress) (const char *procName);
/*!
* This notifies the vendor library which dispatch table index is
* assigned to a particular EGL extension function.
*/
void (*setDispatchIndex) (const char *procName, int index);
/*
* The vendor library may use the isPatchSupported, initiatePatch,
* releasePatch, and patchThreadAttach callbacks to re-write libglvnd's
* entrypoints at make current time, provided no other contexts are current
* and the TLS model supports this functionality. This is a performance
* optimization that may not be available at runtime; the vendor library
* must not depend on this functionality for correctness.
*
* To use this optimization, the vendor library must provide at least the
* isPatchSupported and initiatePatch entrypoints.
*
* Note that these functions are identical to the ones used in GLX.
*/
/*!
* (OPTIONAL) Checks to see if the vendor library supports patching the
* given stub type and size.
*
* \param type The type of entrypoints. This will be a one of the
* __GLDISPATCH_STUB_* values.
* \param stubSize The maximum size of the stub that the vendor library can
* write, in bytes.
* \param lookupStubOffset A callback into libglvnd to look up the address
* of each entrypoint.
*/
GLboolean (* isPatchSupported)(int type, int stubSize);
/*!
* (OPTIONAL) Called by libglvnd to request that a vendor library patch its
* top-level entrypoints.
*
* The vendor library should use the \p lookupStubOffset callback to find
* the addresses of each entrypoint.
*
* This function may be called more than once to patch multiple sets of
* entrypoints. For example, depending on how they're built, libOpenGL.so
* or libGL.so may have their own entrypoints that are separate functions
* from the ones in libGLdispatch.
*
* Note that during this call is the only time that the entrypoints can be
* modified. After the call to \c initiatePatch returns, the vendor library
* should treat the entrypoints as read-only.
*
* \param type The type of entrypoints. This will be a one of the
* __GLDISPATCH_STUB_* values.
* \param stubSize The maximum size of the stub that the vendor library can
* write, in bytes.
* \param lookupStubOffset A callback into libglvnd to look up the address
* of each entrypoint.
*
* \return GL_TRUE if the vendor library supports patching with this type
* and size.
*/
GLboolean (*initiatePatch)(int type,
int stubSize,
DispatchPatchLookupStubOffset lookupStubOffset);
/*!
* (OPTIONAL) Called by libglvnd to notify the current vendor that it no
* longer owns the top-level entrypoints.
*
* Libglvnd will take care of the restoring the entrypoints back to their
* original state. The vendor library must not try to modify them.
*/
void (*releasePatch)(void);
/*!
* (OPTIONAL) Called at the start of window-system functions (GLX and EGL).
* This callback allows vendor libraries to perform any per-thread
* initialization.
*
* This is basically a workaround for broken applications. A lot of apps
* will make one or more invalid GLX/EGL calls on a thread (often including
* a MakeCurrent with invalid parameters), and then will try to call an
* OpenGL function.
*
* A non-libglvnd-based driver would be able to initialize any thread state
* even on a bogus GLX call, but with libglvnd, those calls wouldn't get
* past libGLX.
*
* This function is optional. If it's \c NULL, then libGLdispatch will
* simply ignore it.
*
* \note This function may be called concurrently from multiple threads.
*/
void (*patchThreadAttach)(void);
/*!
* (OPTIONAL) Tries to determine the platform type for a native display.
*
* If the vendor library provides this function, then libglvnd will call it
* to determine which platform to use for a native display handle in
* eglGetDisplay.
*
* If no vendor library identifies the platform, then libglvnd will fall
* back to its own platform detection logic.
*
* Libglvnd can call this function for any native display handle except
* \c EGL_DEFAULT_DISPLAY.
*
* No matter what the value of \p native_display, the vendor library must
* not crash, and must not return a false match. If the vendor library
* can't identify the display, then it must return \c EGL_NONE.
*
* In particular, that means that a vendor library must not return any sort
* of default or fallback platform.
*
* \param native_display The native display handle passed to eglGetDisplay.
* \return Either a platform type enum or EGL_NONE.
*/
EGLenum (* findNativeDisplayPlatform) (void *native_display);
} __EGLapiImports;
/*****************************************************************************/
#define __EGL_MAIN_PROTO_NAME "__egl_Main"
typedef EGLBoolean (* __PFNEGLMAINPROC) (uint32_t version, const __EGLapiExports *exports,
__EGLvendorInfo *vendor, __EGLapiImports *imports);
/*!
* Vendor libraries must export a function called __egl_Main() with the
* following prototype.
*
* This function also performs a handshake based on the ABI version number.
* Vendor libraries can optionally use the version number to support older
* versions of the ABI.
*
* \param[in] version The ABI version. The upper 16 bits contains the major version
* number, and the lower 16 bits contains the minor version number.
*
* \param[in] exports The table of functions provided by libEGL. This pointer will
* remain valid for as long as the vendor is loaded.
*
* \param[in] vendor The opaque pointer used to identify this vendor library. This
* may be used in future versions to provide additional per-vendor information.
*
* \param[out] imports The function table that the vendor library should fill
* in. The vendor library must assign every non-optional function in the
* struct.
*
* \return True on success. If the vendor library does not support the
* requested ABI version or if some other error occurs, then it should return
* False.
*/
EGLBoolean __egl_Main(uint32_t version, const __EGLapiExports *exports,
__EGLvendorInfo *vendor, __EGLapiImports *imports);
/*!
* @}
*/
#if defined(__cplusplus)
}
#endif
#endif /* __LIB_EGL_ABI_H */
+191
View File
@@ -29,6 +29,8 @@ option(MOBILEGL_BUILD_DISAGGREGATED "Build the MG_Remote transport layer (two-pr
# only decides whether the shims are compiled in.
option(MOBILEGL_BUILD_DISAGGREGATED_INPROC "Compile the in-process (one-process, two-role) split shims; implies MOBILEGL_BUILD_DISAGGREGATED" OFF)
option(MOBILEGL_BUILD_SERVER_SPIKE "Build the P0 spike-A MobileGLServer delivery-chain executable (Android only)" OFF)
option(MOBILEGL_BUILD_EXTMEM_PROBE "Package tools/spikes/extmem_probe as libMobileGLExtmemProbe.so for the trace APK's exec hook (P11 A3, Android only)" OFF)
option(MOBILEGL_EXTMEM_PROBE_AS_SERVER "Spike only (P11 B2 step 1): also package the probe as libMobileGLServer.so, the real server as libMobileGLServerReal.so" OFF)
# The PipeInputs strangler (ARCHITECTURE.md 9.2). OFF is the pull build and must stay
# byte-identical to a tree without either option: MGB_CTX is the live GLContext, no
# MGPipe/PipeInputs source is compiled, every MGP_FILL is ((void)0).
@@ -401,7 +403,10 @@ set(SOURCE_FILES
MobileGL/MG_Impl/GLXImpl/LookUp/LookUp.cpp
MobileGL/MG_Impl/EGLImpl/Exporting/Definitions.cpp
MobileGL/MG_Impl/EGLImpl/Exporting/GlvndVendor.cpp
MobileGL/MG_Impl/EGLImpl/EGLImpl.cpp
MobileGL/MG_Impl/EGLImpl/WaylandWindow.cpp
# @INSERTION_POINT:SOURCE_FILE_GLIMPL@ #
MobileGL/MG_Impl/GLImpl/Sampler/Validators.cpp
@@ -432,10 +437,14 @@ set(SOURCE_FILES
MobileGL/MG_Backend/Init.cpp
MobileGL/MG_Backend/BackendObject.cpp
MobileGL/MG_Backend/HostFrameBridge.cpp
MobileGL/MG_Backend/GbmFrameChannel.cpp
MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp
MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp
MobileGL/MG_Backend/DirectGLES/Utils.cpp
MobileGL/MG_Backend/DirectGLES/Managers.cpp
MobileGL/MG_Backend/DirectGLES/HostFrameTarget.cpp
MobileGL/MG_Backend/DirectGLES/MultiDraw.cpp
MobileGL/MG_Backend/DirectVulkan/DirectVulkan.cpp
@@ -592,6 +601,9 @@ endif()
if (MOBILEGL_BUILD_DISAGGREGATED)
message(STATUS "MobileGL: disaggregated transport ON, appending MG_Remote sources")
list(APPEND SOURCE_FILES
MobileGL/MG_Remote/Transport/ShmLink.cpp
MobileGL/MG_Remote/Transport/LinkMetrics.cpp
MobileGL/MG_Remote/Transport/StreamLink.cpp
MobileGL/MG_Remote/Transport/Ring.cpp
MobileGL/MG_Remote/Transport/Doorbell.cpp
MobileGL/MG_Remote/Transport/ShmSegment.cpp
@@ -603,9 +615,18 @@ if (MOBILEGL_BUILD_DISAGGREGATED)
MobileGL/MG_Remote/Transport/FdPassing.cpp
MobileGL/MG_Remote/Transport/InProcessTransport.cpp
MobileGL/MG_Remote/Transport/SocketTransport.cpp
MobileGL/MG_Remote/Transport/ControlInbox.cpp
# P9 W3: the server's log forward - a queue and a sender thread, graded by severity
# (<= WARN lossy, >= ERROR lossless and paced), so no logging thread writes the socket.
MobileGL/MG_Remote/Transport/LogForward.cpp
# Keeps MG_Util/Debug/Log.h - and through it the GL frontend's
# umbrella header - out of the header-only wire code (WireLog.h).
MobileGL/MG_Remote/Transport/WireLog.cpp
# P11 A1: the adopt tier settled at the handshake from the knob and the data plane, by
# both roles (so SERVER-partition code in the link ratchet, not the client's tracker).
MobileGL/MG_Remote/Transport/AdoptTier.cpp
# P11 B2: T0 - the client's AHardwareBuffer store and its hop descriptor (both roles).
MobileGL/MG_Remote/Transport/AdoptT0.cpp
# ---- P5: the three new directories ------------------------------
# Wire/ the G3 codec: MGPWireRec_* in and out of SEG_CMD, blobs and
# var-tails in and out of SEG_STAGE. [w1]
@@ -626,6 +647,7 @@ if (MOBILEGL_BUILD_DISAGGREGATED)
MobileGL/MG_Remote/FatalFunnel.cpp
MobileGL/MG_Remote/Wire/PipeWireCodec.cpp
MobileGL/MG_Remote/Client/ClientSession.cpp
MobileGL/MG_Remote/Client/DmaBufImport.cpp
MobileGL/MG_Remote/Client/EmitTables.cpp
MobileGL/MG_Remote/Client/CapsMirror.cpp
# P5 c1: the client role's BackendObject. pActiveBackendObject holds one of these
@@ -646,11 +668,25 @@ if (MOBILEGL_BUILD_DISAGGREGATED)
MobileGL/MG_Remote/Server/PipeApplier.cpp
MobileGL/MG_Remote/Server/ServerLoop.cpp
MobileGL/MG_Remote/Server/ServerMain.cpp
# P11 PAIR: a socket client's control and aux connections paired by the nonce each
# presents first (PairBind), not by arrival order - both unix supervisors take pairs here.
MobileGL/MG_Remote/Server/PairAcceptor.cpp
# P11 B2: T0's Offers on the aux socket, matched to their map_persistent records by seq.
MobileGL/MG_Remote/Server/AdoptInbox.cpp
MobileGL/MG_Remote/Server/ServerSpawn.cpp
# P5f fc: the value frame that replaced the control mailbox's function-pointer payload,
# and its codec against protocol.fbs's SurfaceOp/SurfaceReply (the spawn arm's seam).
MobileGL/MG_Remote/Server/SurfaceControlFrame.cpp
MobileGL/MG_Remote/Protocol/SurfaceOpCodec.cpp
# P12 (on-screen server window): the process-wide holder of the server's own window
# (the display Activity's SurfaceView on Android; nobody installs one on the host, so a
# ServerOwned request is refused by name there).
MobileGL/MG_Remote/Server/ServerDisplay.cpp
# P7 gate 5 (g5-msprobe): the Magma wire arm's two multisample depth/stencil resolve
# arms and the probe that measures the first against the second, shared by the
# renderer's bring-up and the Vulkan POST. Wire-only - the monolith records neither arm -
# so it stays out of the pull build (gate G1).
MobileGL/MG_Backend/DirectVulkan/Renderer/WireDepthResolveProbe.cpp
)
endif()
@@ -668,6 +704,14 @@ if (ANDROID)
MobileGL/MG_Util/SelfTest/DriverPostJni.cpp
MobileGL/MG_Util/SelfTest/DriverBenchJni.cpp
)
if (MOBILEGL_BUILD_DISAGGREGATED)
# P12 (on-screen server window): the APK's MobileGLDisplayActivity - the in-process
# display server's thread, its SurfaceView's ANativeWindow in and out of ServerDisplay, and
# the geometry up-call. Split build only: nothing else has an in-process server to drive.
list(APPEND SOURCE_FILES
MobileGL/MG_Remote/Server/DisplayServerJni.cpp
)
endif()
endif()
if (WIN32)
@@ -742,6 +786,9 @@ if (MOBILEGL_BUILD_DISAGGREGATED)
# target, and above all no flatc in the build graph. protocol_generated.h
# is committed and regenerated by scripts/gen_protocol.py.
list(APPEND MOBILEGL_INCLUDE_DIR ${CMAKE_SOURCE_DIR}/3rdparty/flatbuffers/include)
# The vendored glvnd vendor-ABI headers (3rdparty/glvnd): the entry point a vendor library
# exposes so the system libEGL - which IS glvnd - can dispatch to it.
list(APPEND MOBILEGL_INCLUDE_DIR ${CMAKE_SOURCE_DIR}/3rdparty/glvnd/include)
endif()
add_library(${CMAKE_PROJECT_NAME} SHARED
@@ -779,6 +826,17 @@ target_link_libraries(${CMAKE_PROJECT_NAME}
${MOBILEGL_LINK_LIBRARIES}
)
# A LINUX BUILD COMPILES THE DirectVulkan BACKEND BUT HAS NO VULKAN LOADER TO LINK AGAINST, so the
# resulting shared library carries undefined vk* symbols - and that is not theoretical: glvnd dlopens
# this file as the EGL vendor with RTLD_NOW, so the load fails outright with "undefined symbol:
# vkGetQueryPoolResults" (measured on the phone's container). Android links the loader already; on
# Linux the caller passes the runtime's own libvulkan.so.1, and leaving the variable empty keeps the
# previous behaviour for every build that has no Vulkan runtime to point at.
set(MOBILEGL_VULKAN_LOADER "" CACHE FILEPATH "Vulkan loader (libvulkan.so.1) to link into the GL library")
if (UNIX AND NOT ANDROID AND MOBILEGL_VULKAN_LOADER)
target_link_libraries(${CMAKE_PROJECT_NAME} PRIVATE ${MOBILEGL_VULKAN_LOADER})
endif()
target_compile_definitions(${CMAKE_PROJECT_NAME}
PUBLIC
${MOBILEGL_COMPILE_DEF}
@@ -845,6 +903,68 @@ if(NOT ANDROID)
)
endif()
# ---------------------------------------------------------------------------
# MobileGLGbm: the GBM drop-in the container's compositor loads.
#
# kwin - the Wayland compositor in the container half of the split - allocates the
# buffers it renders into and scans out through libgbm, and the dma-buf fds it gets
# back are what the Anland display producer ships to the phone. Mesa's libgbm cannot
# serve that: its buffers belong to the host GPU, and the phone would receive a copy
# through a path that was never designed to carry one. This target is the library
# that is loaded in its place, under the name a loader looks for.
#
# It is deliberately NOT part of libMobileGL.so. The compositor needs one file named
# libgbm.so that it loads instead of Mesa's, and folding this code into the GL
# library would put a second GL frontend, a second Vulkan instance and a second
# glslang in the compositor's address space to serve an API that shares none of them.
#
# The include paths below are borrowed rather than linked. Includes.h - which this
# library includes for the MobileGL types and the MGLOG macros - pulls in the GL,
# EGL, Vulkan, glslang, SPIRV-Cross and xxHash HEADERS transitively, and headers are
# all this library wants from them: linking any of those static libraries into a
# libgbm would embed a copy of each in a process that already has one.
#
# -Bsymbolic-functions, for the same reason the main target carries it: the point of
# this library is to REPLACE a libgbm that may already be loaded in the process, so
# every internal reference has to bind to this file's own definitions rather than to
# whatever else the process has under the same names.
# ---------------------------------------------------------------------------
add_library(MobileGLGbm SHARED
MobileGL/MG_Gbm/GbmDevice.cpp
# The display host's socket protocol, kept in its own translation unit so that the
# allocation paths never have to know how a frame arrives - only that one did.
MobileGL/MG_Gbm/GbmHostChannel.cpp
MobileGL/MG_Gbm/GbmApi.cpp
# MGLOG_I/W/E expand to a call to MobileGL::MG_Util::Debug::Log, and this is the
# one translation unit that defines it. It is compiled in rather than linked from
# libMobileGL.so because this library has to load into a compositor that has no
# libMobileGL in it at all - and because a libgbm with an undefined symbol is a
# library that fails on the first diagnostic rather than at a link step somebody
# could have read.
MobileGL/MG_Util/Debug/Log.cpp
)
# The vendored gbm.h first: this library implements THAT declaration set, and a
# system <gbm.h> from another Mesa version on the include path ahead of it would
# change the ABI under the implementations in GbmApi.cpp.
target_include_directories(MobileGLGbm PRIVATE
${CMAKE_SOURCE_DIR}/3rdparty/gbm/include
${MOBILEGL_INCLUDE_DIR}
${CMAKE_SOURCE_DIR}/3rdparty/glslang
${CMAKE_SOURCE_DIR}/3rdparty/SPIRV-Cross
${CMAKE_SOURCE_DIR}/3rdparty/xxHash
${CMAKE_SOURCE_DIR}/3rdparty/Vulkan-Headers/include
)
set_target_properties(MobileGLGbm PROPERTIES
OUTPUT_NAME gbm
POSITION_INDEPENDENT_CODE ON
)
if (UNIX AND NOT APPLE)
target_link_options(MobileGLGbm PRIVATE "LINKER:-Bsymbolic-functions")
endif()
if (TRACY_ENABLE)
target_link_libraries(${CMAKE_PROJECT_NAME} PUBLIC Tracy::TracyClient)
target_link_libraries(${CMAKE_PROJECT_NAME}_s PUBLIC Tracy::TracyClient)
@@ -852,6 +972,19 @@ if (TRACY_ENABLE)
target_compile_definitions(${CMAKE_PROJECT_NAME}_s PUBLIC -DTRACY_ENABLE)
endif ()
# GLVND, AND WHY THIS IS A LINKER FLAG RATHER THAN A VISIBILITY ONE.
#
# MobileGL can serve as a glvnd EGL vendor - the library the system's libEGL.so.1 dispatches to -
# and a vendor must hand glvnd ITS OWN entry points: glvnd calls whatever address it is given, so
# an address that resolves to another libEGL in the process is a call back into the dispatcher.
# With default visibility every reference inside this library resolves against the whole process,
# and a process that loaded the system libEGL first wins. Marking the entry points protected says
# the same thing in the source, except the platform's own GL/EGL headers declare them default and
# the two visibilities do not mix; this flag is that promise made by the linker instead.
if (UNIX AND NOT APPLE)
target_link_options(${CMAKE_PROJECT_NAME} PRIVATE "LINKER:-Bsymbolic-functions")
endif()
if (ANDROID)
target_link_libraries(${CMAKE_PROJECT_NAME} PUBLIC
android
@@ -961,6 +1094,12 @@ if (ANDROID AND MOBILEGL_BUILD_INTEGRATION_TEST)
add_subdirectory(MobileGL/MG_IntegrationTest)
endif()
# The replay runner also builds for Android: an adb-shell executable that can
# dial a render server on the device (unix or TCP) without any Activity.
if (ANDROID AND MOBILEGL_BUILD_TRACE_REPLAY)
add_subdirectory(tools/trace_replay)
endif()
# ---------------------------------------------------------------------------
# P0 spike A: the Android delivery chain for a second native executable.
#
@@ -1069,3 +1208,55 @@ if (ANDROID AND MOBILEGL_BUILD_SERVER_SPIKE)
RUNTIME_OUTPUT_DIRECTORY "${CMAKE_LIBRARY_OUTPUT_DIRECTORY}")
endif()
endif()
if (ANDROID AND MOBILEGL_BUILD_EXTMEM_PROBE)
# P11 A3 (open question 3): the spike-B external-memory probe, exec'd from the trace
# app's own process by the spike hook (`--es mobilegl_spike_spawn
# libMobileGLExtmemProbe.so`), so its rows answer for the untrusted_app domain. The
# same PIE-executable-named-like-a-library shape as MobileGLServerSpike above; it
# links nothing from MobileGL.
add_executable(MobileGLExtmemProbe
${CMAKE_CURRENT_SOURCE_DIR}/tools/spikes/extmem_probe/extmem_probe.cpp)
set_target_properties(MobileGLExtmemProbe PROPERTIES
PREFIX "lib"
SUFFIX ".so"
OUTPUT_NAME "MobileGLExtmemProbe"
POSITION_INDEPENDENT_CODE ON)
target_compile_options(MobileGLExtmemProbe PRIVATE -fPIE)
target_link_options(MobileGLExtmemProbe PRIVATE -pie)
target_link_libraries(MobileGLExtmemProbe PRIVATE vulkan EGL GLESv3 android log)
if (CMAKE_LIBRARY_OUTPUT_DIRECTORY)
set_target_properties(MobileGLExtmemProbe PROPERTIES
RUNTIME_OUTPUT_DIRECTORY "${CMAKE_LIBRARY_OUTPUT_DIRECTORY}")
endif()
if (MOBILEGL_EXTMEM_PROBE_AS_SERVER)
# P11 B2 step 1, spike APKs only: MobileGLServerService execs `libMobileGLServer.so
# <endpoint> --serve` and B1's broker connects to whatever that listens on, so the probe
# takes the server's file name and answers "does a client's AHardwareBuffer cross this
# route into the server app" on the real route (tools/spikes/extmem_probe, route mode).
if (TARGET MobileGLServer)
set_target_properties(MobileGLServer PROPERTIES OUTPUT_NAME "MobileGLServerReal")
endif()
add_executable(MobileGLExtmemProbeAsServer
${CMAKE_CURRENT_SOURCE_DIR}/tools/spikes/extmem_probe/extmem_probe.cpp)
set_target_properties(MobileGLExtmemProbeAsServer PROPERTIES
PREFIX "lib"
SUFFIX ".so"
OUTPUT_NAME "MobileGLServer"
POSITION_INDEPENDENT_CODE ON)
target_compile_options(MobileGLExtmemProbeAsServer PRIVATE -fPIE)
target_link_options(MobileGLExtmemProbeAsServer PRIVATE -pie)
target_link_libraries(MobileGLExtmemProbeAsServer PRIVATE vulkan EGL GLESv3 android log)
if (CMAKE_LIBRARY_OUTPUT_DIRECTORY)
set_target_properties(MobileGLExtmemProbeAsServer PROPERTIES
RUNTIME_OUTPUT_DIRECTORY "${CMAKE_LIBRARY_OUTPUT_DIRECTORY}")
endif()
endif()
endif()
# The GBM library carries the tree's logger, which reaches Android's own log. The main target gets
# these libraries in its own Android block; a library built from the same sources needs them for the
# same reason, and without them it is the LINK that fails rather than anything in the code.
if (ANDROID)
target_link_libraries(MobileGLGbm PRIVATE log android)
endif()
+188 -8
View File
@@ -132,6 +132,18 @@ namespace MobileGL::MG_Config {
Bool MagmaR11G11B10FFallback = false;
// MOBILEGL_MAGMA_FRAMESINFLIGHT: requested Magma frames in flight, defaulting to 3.
Uint32 MagmaFramesInFlight = 3;
// MOBILEGL_MAGMA_MAX_DRAWS_PER_COMMAND_BUFFER: GL draws and dispatches Magma records into
// one command buffer before it submits it and continues on a fresh one. 0 = unbounded.
// Drivers back a command buffer with GPU memory that is only returned when the buffer is
// freed; Adreno maps it into the process in 16 KiB chunks, so one ~1M-draw loading frame
// exhausts vm.max_map_count. See VulkanRenderer::SplitOversizedRecording.
Uint32 MagmaMaxDrawsPerCommandBuffer = 16384;
// MOBILEGL_MAGMA_DESCRIPTOR_TRIM_FRAMES: how many consecutive frames (strictly, BeginFrame
// epochs) a frame slot's descriptor pools must stay under a quarter full before Magma
// hands the slot's whole set cache back and lets it re-grow. 0 = never trim. A loading
// frame otherwise keeps its burst of sets (rd12: ~15k, where steady frames use a few dozen)
// for the life of the process. See UniformManager::TrimFrameDescriptorPools.
Uint32 MagmaDescriptorTrimFrames = 120;
// MOBILEGL_ESPRYT_AVOID_SAMPLER_MIPMAP_MIN_FILTER: avoid mipmap min filters in samplers,
// resolves certain rendering bugs on ANGLE + llvmpipe.
Bool EsprytAvoidSamplerMipmapMinFilter = false;
@@ -202,9 +214,12 @@ namespace MobileGL::MG_Config {
// persistently+coherently mapped storage at definition time (negative control /
// escape hatch). Frontend-scoped: it engages only where the active backend
// provides AcquirePersistentMap. With adoption on, an app SubData into a busy
// 128MB arena is a plain memcpy into GPU-visible memory; every driver-mediated
// route for the same write stalls the thread or ghost-copies the whole arena on
// this class of Mali driver, and the arena stops costing its size again in RAM.
// 128MB arena lands GPU-ordered through the backend's resident-subdata op
// (BufferObject::LandBytesIntoResidentStore, since 0ee3384b: an in-place write
// tore the frames still reading the old bytes) - still no driver-mediated write
// into the busy mutable store, which on this class of Mali driver stalls the
// thread or ghost-copies the whole arena, and the arena stops costing its size
// again in RAM.
Bool DisableLargeBufferAdoption = false;
// MOBILEGL_ESPRYT_FORCE_DS_READBACK_EMULATION: make DirectGLES skip the native ES
// depth/stencil reads and always go through the shader-sampling emulation. Core GL
@@ -456,6 +471,20 @@ namespace MobileGL::MG_Config {
// The <path> of `unix:` / the <name> of `pipe:`. Empty for the other three modes.
extern String TransportEndpoint;
// P7 F1. DID **THIS PROCESS'S OWN CONFIGURATION** ASK FOR A SPLIT TRANSPORT.
//
// `Transport` alone cannot answer that, and the difference is exactly the one the F1 gate
// turns on. A unit fixture assigns `Transport` by hand to put the code under test on its
// split arm (ServerLoopTest's main does, and says why) while building no client at all; a
// real run gets it from MG_ConfigLoader::Init(), and that run WILL bring a client half up.
// Only the second may be held to "never decide a record family's fate from a placeholder
// caps mirror" - in the first there is no handshake, no mirror to adopt and nothing the
// rule could mean.
//
// Written once, by InitTransport(), and never cleared: a process does not stop having been
// configured. Read by MG_Remote::Client::CapsMirror::RequireFirstSnapshot.
extern Bool SplitTransportRequestedByConfig;
// The MOBILEGL_IPC_* family. A separate table rather than more FeaturesTable members,
// for the G1 reason above and because every field here is meaningless without the
// transport: a build that cannot reach the MG_Remote code cannot honour one of them.
@@ -465,12 +494,25 @@ namespace MobileGL::MG_Config {
// the full planned inventory (PRESENT_CREDIT, POLL_ESCALATE, SHADOW_SHM,
// INLINE_PAYLOADS, TRACE, ATTACH, RESPAWN, IDLE_EXIT_S), and every one of those belongs
// to P6 or later.
//
// P12 (on-screen server window): MOBILEGL_IPC_SURFACE's two values. See IpcTable::Surface.
enum class IpcSurface : Uint8 {
Offscreen = 0, // a window surface is the client's own window, as before (the default)
Server = 1, // a window surface is the SERVER's window: WindowKind::ServerOwned
Host = 2, // a window surface draws into frames the DISPLAY HOST owns; the host presents
};
// IpcTable::AdoptTier when MOBILEGL_IPC_ADOPT_TIER is unset (or unparseable): T0 by default,
// with QUIET fallbacks. Not a tier a user can type - the knob admits 0..2.
inline constexpr Uint32 kAdoptTierUnset = 3;
struct IpcTable {
// MOBILEGL_IPC_SERVER_PATH: where to find libMobileGLServer. P6 consumes it; P5
// lands the parse because t1's ctest ENVIRONMENT blocks and add_trace_replay_test's
// SPLIT variant already carry it, and an environment variable that nothing parses is
// indistinguishable from one that is parsed and ignored.
String ServerPath;
// Control endpoint and data-plane selection are independent of topology.
String Control = "fork"; // fork | unix:<path> | tcp://host:port
String Data = "auto"; // auto | shm | stream
// MOBILEGL_IPC_RING_MB: SEG_CMD size. A RECORD MAY BE AT MOST HALF OF THIS
// (RingProducer::MaxRecordBytes), so 8 MiB caps one record at 4 MiB; R-10 makes the
// codec publish a max-record-bytes counter rather than assume that is enough.
@@ -480,8 +522,37 @@ namespace MobileGL::MG_Config {
// stage chunk budget (MGPipeStageChunkBytes, a quarter of this), and a record type with
// no cut is Fatal{RingOverrun, "SEG_STAGE"} rather than allowed to exceed it.
Uint32 StageMb = 32;
// MOBILEGL_IPC_WIRE_DEFERRED_MB (P7 wave 4 M2, ID-P7-32): the SERVER's budget, in MiB,
// for orphaned wire buffer stores - the old VkBuffer every glBufferData that crosses the
// wire leaves behind - that a GPU command recorded but not yet retired may still name.
// Stores no command names are destroyed at once and stores whose last submission has
// retired are destroyed at the next park; this bounds the REST. When the parked bytes
// exceed it after a sweep, the server flushes what it has recorded and waits for it
// (WaitForWireBufferHostAccess's sync point, mid-frame), which retires every one. It is
// not a frame count because a frame is not bounded: a snapshot-exiting pbuffer replay
// delivers one present for 1.3 M calls. The same sync point also fires above a fixed
// 1024 parked stores (VkBufferManager::kWireDeferredCountCeiling), because small
// orphans never reach a byte budget. 0 IS THE NEGATIVE CONTROL for both, not "unlimited
// by design": no forced sync, so a one-frame respecify-and-draw loop grows without bound
// and MagmaWireReclaimScenario's watermark cases must go red.
Uint32 WireDeferredMb = 64;
// MOBILEGL_IPC_SPIN_US: spin before parking on a doorbell, either direction.
Uint32 SpinUs = 50;
// MOBILEGL_IPC_EVENT_WAIT_MS (PH-6, ID-P7-2): the SERVER's patience, in ms, for ONE
// reverse-channel event that finds SEG_EVENT full under run-ahead. The apply thread
// publishes, rings the client and parks until the client drains enough room; a client
// that has not made room when this runs out - it stopped draining (NotDraining) or it
// drains a slot at a time (TooSlow) - FORFEITS the reverse channel: that event and every
// later one is dropped and counted (eventDropped), and the session stops by the ordinary
// stop path rather than by Fatal{EventRingOverflow}. A peer that goes away (PeerGone) or
// a session that is stopped (Stopped) ends the wait at once instead, so the knob may
// exceed ServerLoop::Stop()'s 5000 ms join. It bounds the whole reservation, not one
// park, so a trickling peer cannot stretch it - nor can a shm peer that writes the
// (peer-writable) eventRingFull latch to 0 itself, though that one keeps the apply
// thread spinning rather than parked until the budget runs out. Was a 30000 ms constant
// spent twice. Read by the server only (ServerSpawn.cpp passes it to a spawned child);
// the lockstep arm (no kCapRunAheadApply, i.e. Magma) never waits and keeps P5C's Fatal.
Uint32 EventWaitMs = 2000;
// MOBILEGL_IPC_PERSISTENT_BLOCK_KB: block granularity of the persistent-map push.
// 0 IS A NEGATIVE CONTROL, NOT "unlimited": it disables the push, and
// PersistentCoherentMapScenario must go RED under it (exit gate E3(a)).
@@ -496,11 +567,80 @@ namespace MobileGL::MG_Config {
// every reply-slot row), which is the only place BARRIER-PULLED fields are read. 0
// restores the per-record barrier of R-1.
Uint32 BatchWaits = 1;
// MOBILEGL_IPC_ADOPT_TIER: 2 = emulate (client keeps the shadow and pushes), which
// is the only tier P5 implements and the reason persistent-map-push can be non-zero
// at all (R-6). 0 and 1 parse and are Fatal at use with "P11"; they exist now so the
// negative control has a spelling the day P11 writes it.
Uint32 AdoptTier = 2;
// MOBILEGL_IPC_CREATE_WINDOW: how many RESOURCE_CREATE answers may be outstanding before the
// client waits for one. 1 = every create waits (the shape R-5 shipped with); N > 1 = up to N
// creates go out fire-and-forget, their caller reads a PROVISIONAL accept, and the oldest
// answer is taken when the window fills.
//
// DEFAULT 2, AND IT TOOK THREE THINGS TO GET HERE. The deferral was inert for two
// measured reasons and one structural one, in the order they were found:
// 1. the reply buffer was a ring of the last 8 answers, and between two creates this
// client emits on the order of a thousand records - so the answer was always gone
// (`want=310` on every miss while the ring spanned [1175..1504]);
// 2. a residency budget that tried to take answers back early could not help, because
// what displaces an answer is what the SERVER sends, not what the client emits;
// 3. and so the fix was RETENTION - a link keeps a declared answer until its reader
// takes it (Transport/ILink.h's RetainsReplies, StreamLink's declared-answer store).
//
// MEASURED ON THE DEVICE, load frame, both ends rebuilt (create 4,536, records 103,097,
// frame-sent identical per op throughout):
// window 1 4,536 blocking takes 28.66 s (the shape R-5 shipped)
// window 2 2,500 17.27 s
// window 4 1,644 11.02 s
// and the round trip does NOT grow as the window deepens (5,285 / 4,082 us against 5,672
// at 1), which is the measurement that says the ~5 ms is a WAKEUP the server batches away
// rather than its apply throughput - so amortising four creates per wakeup is a real 2.6x.
//
// 6 IS THE DEFAULT, AND IT IS WHERE THE CURVE FLATTENS - the sweep that says so is the
// one the paragraph above asked for, and it moved the ceiling too (kCreateWindowMax is 8
// now; the old 4 was derived from a reply pool the window does not run on). Three
// interleaved runs per arm on one binary, load frame, create 4,536 sent in every arm:
//
// window 4 1,503 blocking takes 13.23 s median (spread 11.37-16.29)
// window 5 1,277 11.31 s (10.61-12.12)
// window 6 1,155 10.79 s ( 9.69-10.81)
// window 8 859 10.96 s ( 8.92-11.03)
//
// 4->5 buys 1.92 s, 5->6 buys 0.53 s, 6->8 buys NOTHING (-0.18 s, inside a session noise
// of ~1.5-2 s) - so 6 is the smallest window on the flat part and the extra answers a
// deeper one holds buy no time. AND PAST THIS POINT THE FRAME IS NOT WAIT-BOUND ANY MORE:
// window 8 pays 296 fewer blocking takes than window 6 and the wall clock does not move.
// The store is not the constraint either - its peak is logged as it crosses (StreamLink's
// P12 line) and reads 6..11 against a cap of 24 at every arm from 5 up. What is left is
// the constant ~350-400 creates that block for a reason depth does not reach, and the
// 101 MB / 103k records the SERVER has to apply. NOTE THE PREDICATE the whole mechanism rests
// on: an answer is retained because the wire DECLARED it will be read - NOT because its
// caller was fire-and-forget. The window's own creates are fire-and-forget AND read, so a
// flag keyed on wantReply would disable the very mechanism it exists to enable
// (docs/Disaggregated/notes/p12/CREATE-WINDOW-MEASURED.md sections 7-11).
Uint32 CreateWindow = 10;
// MOBILEGL_IPC_PBO_READBACK_SYNC (P9 W1, MG_Remote/CONTRACT-P9.md §1): 1 = a glReadPixels
// or glGetTexImage into a bound GL_PIXEL_PACK_BUFFER takes the reply-slot form again -
// the client waits for the pixels and uploads them into the buffer itself, as before P9.
// 0 (default) sends read_pixels_to_buffer / get_texture_image_to_buffer, waits for
// nothing, and marks the buffer GPU-written. The 1 arm is the A/B control and the
// red-once of the wait-count gate (PackBufferReadbackScenario): it must show a reply
// wait per read.
Uint32 PboReadbackSync = 0;
// MOBILEGL_IPC_POLL_ESCALATE (P10, MG_Remote/CONTRACT-P10.md §1): a glClientWaitSync with a
// zero timeout or a glGetSynciv(GL_SYNC_STATUS) is answered by the client from what the
// server reported (kEventFenceSignaled) - "signaled" when it has been, "not yet" otherwise
// - and only the Nth "not yet" in a row on one fence crosses as a real round trip, so a
// poll loop cannot outlive a late report. 0 = every poll crosses, as before P10: the A/B
// arm and the red-once of the no-reply gate (FencePollScenario).
Uint32 PollEscalate = 64;
// MOBILEGL_IPC_ADOPT_TIER, THE ZERO-COPY SWITCH (ruling ID-P11-14, CONTRACT-P11 B2):
// UNSET (kAdoptTierUnset, the default since 2026-09-29, the user's decision) = T0 where the
// session can run it, T2 QUIETLY where it cannot (MGLOG_D only, still counted);
// 0 = T0 asked by name - a T0 this session cannot use (a stream plane, the server's allow
// switch, no AHardwareBuffer, a failed server POST) is T2 plus one named line;
// 1 = T1, closed - refused by name, then T2; 2 = emulate (T2: the client keeps the shadow
// and pushes). Never a Fatal (MG_Remote/Transport/AdoptTier.h).
Uint32 AdoptTier = kAdoptTierUnset;
// MOBILEGL_IPC_ALLOW_ADOPT_T0 (server): 1 (default) = grant a client's T0 ask when this
// platform and the backend's POST self-test allow it; 0 = never grant it (the session
// runs T2 and says so). Read by the server role only; kept through ServerSpawn.
Uint32 AllowAdoptT0 = 1;
// MOBILEGL_IPC_VERB_BARRIER: 1 = the client blocks at every verb boundary until
// appliedSeq reaches its emitSeq (R-1). 0 is the negative control: it is EXPECTED to
// be red, because 31 of the 63 PipeInputs fields are still pulled from a live
@@ -533,6 +673,21 @@ namespace MobileGL::MG_Config {
// client that is already CPU-bound and costs a frame of latency, which is why the
// default is 1 and not "as deep as the ring".
Uint32 PresentCredit = 1;
// MOBILEGL_IPC_CONTROL_TIMEOUT_MS (CONTRACT-P6 §5.4 D5b): how long a spawn/tcp client
// waits for the SurfaceReply to one surface-control op (eglCreate*Surface, MakeCurrent,
// ...) once the server's backend is up. Its expiry is NOT fatal: the doorbell's death
// latch decides dead (device lost) from alive-but-silent (a named diagnostic). The
// contract named this knob from P6 on; the client hard-coded its default until P7.
Uint32 ControlTimeoutMs = 5000;
// MOBILEGL_IPC_COLD_START_MS (P7): the same wait for the three ops a server may bring its
// NATIVE backend up inside - CreatePbufferSurface, CreateWindowSurface, MakeCurrent -
// until the session's first MakeCurrent is answered ok. The bring-up is lazy (Espryt's
// eglInitialize, Magma's Vulkan instance and device), ~100 ms on a workstation and more
// than the steady 5 s on a loaded CI runner (retrace-split spawn legs, runs 35671704873
// and 35706183230). Never shorter than CONTROL_TIMEOUT_MS; its expiry is the same
// non-fatal named answer. The default is the spawn connect budget, for the same reason:
// "the server has to create a backend, and a cold software rasteriser is not fast".
Uint32 ColdStartMs = 20000;
// MOBILEGL_IPC_STRICT_ERRORS: promote a BARRIER-PULLED field read - and, in a split
// build, the seven sticky forwards that are otherwise exempt - from "count it in
// rsp" to Fatal (R-7.3).
@@ -557,8 +712,33 @@ namespace MobileGL::MG_Config {
// because the resolved mask is logged by whoever starts the apply thread, and the
// string is what an operator typed.
String ServerAffinity = "auto";
// MOBILEGL_IPC_SURFACE (P12, on-screen server window) = offscreen | server. `server` makes
// the client a HEADLESS one: eglCreateWindowSurface / eglCreatePlatformWindowSurface accept
// any native window including NULL, send ONE CreateWindowSurface naming
// WindowKind::ServerOwned (token 0, the size from EGL_WIDTH/EGL_HEIGHT, 0/0 = the server's
// own) and no SetWindowHandle, and take the surface's real geometry back from the server.
// Meaningful only when this client talks to a remote server (MOBILEGL_TRANSPORT=spawn, the
// fork or tcp:// control); under monolith / inproc it is parsed, logged as ignored, and
// changes nothing. `offscreen` (the default) is today's behaviour byte for byte.
// Pbuffers stay pbuffers in both modes (D4: one mode per session, decided by the first).
IpcSurface Surface = IpcSurface::Offscreen;
};
extern IpcTable Ipc;
// P12: the one predicate every client-side consumer of MOBILEGL_IPC_SURFACE asks - a window
// surface is the server's when the knob says so AND there is a remote server to own it.
inline Bool ServerOwnedWindowSurfaces() {
return Transport == TransportMode::Spawn && Ipc.Surface == IpcSurface::Server;
}
// P12: the other on-screen shape, and the mirror image of the one above. The display host
// (the Android side that owns the screen) allocates the frames, offers each one over the
// host frame bridge, and puts it on the glass when this side says it is drawn - so a
// client that asks for this has no window of its own either, and the server takes the
// frames rather than making a window.
inline Bool HostFrameWindowSurfaces() {
return Transport == TransportMode::Spawn && Ipc.Surface == IpcSurface::Host;
}
#else
// The whole point: in a build without MG_Remote this folds at compile time, so
// `if (MG_Config::Transport != MG_Config::TransportMode::Monolith)` in Init.cpp is a
+100 -9
View File
@@ -16,6 +16,11 @@
#include <cerrno>
#include <cstdlib>
#if MOBILEGL_BUILD_DISAGGREGATED
// std::strcmp, for InitIpc's server-role check (P12). Split-only, so the pull build's TU is
// unchanged (G1).
#include <cstring>
#endif
#ifndef _WIN32
extern char** environ;
@@ -32,6 +37,7 @@ namespace MobileGL::MG_Config {
// and there is nothing to define.
TransportMode Transport = TransportMode::Monolith;
String TransportEndpoint;
Bool SplitTransportRequestedByConfig = false;
IpcTable Ipc;
#endif
} // namespace MobileGL::MG_Config
@@ -224,6 +230,9 @@ namespace MobileGL::MG_ConfigLoader {
features.AdvertiseFp64 = QueryEnvFlag("MOBILEGL_ADVERTISE_FP64");
features.MagmaR11G11B10FFallback = QueryEnvFlag("MOBILEGL_MAGMA_R11G11B10F_FALLBACK");
features.MagmaFramesInFlight = QueryEnvUint32("MOBILEGL_MAGMA_FRAMESINFLIGHT", 3, 1, 64);
features.MagmaMaxDrawsPerCommandBuffer =
QueryEnvUint32("MOBILEGL_MAGMA_MAX_DRAWS_PER_COMMAND_BUFFER", 16384, 0, 1u << 24);
features.MagmaDescriptorTrimFrames = QueryEnvUint32("MOBILEGL_MAGMA_DESCRIPTOR_TRIM_FRAMES", 120, 0, 1u << 20);
features.EsprytAvoidSamplerMipmapMinFilter =
QueryEnvFlag("MOBILEGL_ESPRYT_AVOID_SAMPLER_MIPMAP_MIN_FILTER");
features.EsprytAvoidExplicitLodBias = QueryEnvFlag("MOBILEGL_ESPRYT_AVOID_EXPLICIT_LOD_BIAS");
@@ -330,6 +339,10 @@ namespace MobileGL::MG_ConfigLoader {
}
if (lowered == "inproc") {
MG_Config::Transport = MG_Config::TransportMode::InProcess;
// P7 F1: see Config.h. Set beside the mode, on the two arms that actually resolve
// one, and NOT on the two that name a transport and then stay monolith - a process
// that was refused its transport brings no client half up either.
MG_Config::SplitTransportRequestedByConfig = true;
MGLOG_I("Config: MOBILEGL_TRANSPORT=inproc - the MGPipe record stream crosses a real "
"ring to an apply thread");
return;
@@ -342,6 +355,7 @@ namespace MobileGL::MG_ConfigLoader {
// rather than the fork-coupled shape ARCHITECTURE.md:488 described.
if (lowered == "spawn") {
MG_Config::Transport = MG_Config::TransportMode::Spawn;
MG_Config::SplitTransportRequestedByConfig = true; // P7 F1, as above.
// THE SAME SHAPE AS THE inproc LINE ABOVE, AND FOR THE SAME REASON.
// run_trace_case.cmake asserts on a distinctive sentence from this
// function rather than on `MOBILEGL_TRANSPORT=spawn` alone, because
@@ -378,13 +392,22 @@ namespace MobileGL::MG_ConfigLoader {
inline void InitIpc() {
auto& ipc = MG_Config::Ipc;
QueryEnvVariable("MOBILEGL_IPC_SERVER_PATH", ipc.ServerPath, "");
QueryEnvVariable("MOBILEGL_IPC_CONTROL", ipc.Control, "fork");
QueryEnvVariable("MOBILEGL_IPC_DATA", ipc.Data, "auto");
// Both ring floors are 1 MiB, not 0: a ring caps ONE record at half its size, and
// the catalogue's largest fixed payload (MGPFramebufferState, 304 bytes) plus a
// create_shader_state archive already needs far more than a toy ring. The ceilings
// are sanity, not policy.
ipc.RingMb = QueryEnvUint32("MOBILEGL_IPC_RING_MB", 8, 1, 1024);
ipc.StageMb = QueryEnvUint32("MOBILEGL_IPC_STAGE_MB", 32, 1, 4096);
// P7 wave 4 M2 (Config.h has the semantics). 0 is admitted ON PURPOSE as the negative
// control: no forced sync, and MagmaWireReclaimScenario's watermark case must go red.
ipc.WireDeferredMb = QueryEnvUint32("MOBILEGL_IPC_WIRE_DEFERRED_MB", 64, 0, 65536);
ipc.SpinUs = QueryEnvUint32("MOBILEGL_IPC_SPIN_US", 50, 0, 1000000);
// PH-6 (ID-P7-2; Config.h has the semantics). 0 is NOT admitted: a server with no
// patience at all would forfeit a healthy run-ahead client the first time its ring
// filled between two of the client's drains, which is an ordinary backlog.
ipc.EventWaitMs = QueryEnvUint32("MOBILEGL_IPC_EVENT_WAIT_MS", 2000, 1, 600000);
// 0 is admitted ON PURPOSE and is the negative control of exit gate E3(a): it turns
// the persistent-map push OFF, and PersistentCoherentMapScenario must go red.
ipc.PersistentBlockKb = QueryEnvUint32("MOBILEGL_IPC_PERSISTENT_BLOCK_KB", 64, 0, 65536);
@@ -392,14 +415,30 @@ namespace MobileGL::MG_ConfigLoader {
// xxHash64 changed since the last push are shipped.
ipc.PersistentHashSuppress = QueryEnvUint32("MOBILEGL_IPC_PERSISTENT_HASH_SUPPRESS", 1, 0, 1);
ipc.BatchWaits = QueryEnvUint32("MOBILEGL_IPC_BATCH_WAITS", 1, 0, 1);
// The create window. DEFAULT 6, and it works because a link that declares RetainsReplies
// keeps a declared answer until its reader takes it - Config.h carries the three reasons it
// was inert before, the device numbers, and the sweep that put the default at 6 rather
// than 4 or 8 (the curve flattens at 6: 4->5 buys 1.92 s, 5->6 0.53 s, 6->8 nothing).
// Capped at 8: the retained set is bounded by the window's own depth, the store holds 24
// declared answers, and the device sweep showed the wall clock still falling at the old
// cap of 4 (see WireTables.cpp's kCreateWindowMax). WireTables.cpp's static_assert and
// StreamLink's crossing log are the checks on this number.
ipc.CreateWindow = QueryEnvUint32("MOBILEGL_IPC_CREATE_WINDOW", 10, 1, 12);
// P9 W1: the pack-buffer readback's A/B and red-once control (Config.h says both).
ipc.PboReadbackSync = QueryEnvUint32("MOBILEGL_IPC_PBO_READBACK_SYNC", 0, 0, 1);
// P10: fence polls answered locally; 0 is the A/B arm (Config.h).
ipc.PollEscalate = QueryEnvUint32("MOBILEGL_IPC_POLL_ESCALATE", 64, 0, 1000000);
// The verify harness compares the pushed block against the applier per verb; a
// batched queue lets the comparer read a supplied field mid-apply, which is a
// torn read rather than a divergence. The batch is therefore off whenever the
// shadow comparer is armed.
if (MG_Config::Features.PipeVerify) ipc.BatchWaits = 0;
// 2 is the only tier P5 implements (R-6). 0 and 1 parse here and are refused at the
// point of use, which is where the "P11" in the message belongs.
ipc.AdoptTier = QueryEnvUint32("MOBILEGL_IPC_ADOPT_TIER", 2, 0, 2);
// UNSET is T0 with quiet fallbacks (the default since 2026-09-29); 0 asks for T0 by name, 1 is
// refused by name, 2 is T2 - settled at the handshake from the data plane (P11 A1/B2,
// MG_Remote/Transport/AdoptTier.h). An unparseable value is warned about and reads as unset.
ipc.AdoptTier = QueryEnvUint32("MOBILEGL_IPC_ADOPT_TIER", MG_Config::kAdoptTierUnset, 0, 2);
// P11 B2: the server's own T0 allow switch (Config.h).
ipc.AllowAdoptT0 = QueryEnvUint32("MOBILEGL_IPC_ALLOW_ADOPT_T0", 1, 0, 1);
ipc.VerbBarrier = QueryEnvUint32("MOBILEGL_IPC_VERB_BARRIER", 1, 0, 1);
// P5e (MG_Remote/CONTRACT-P5E.md §1). The wait rule's A/B, parsed here like every
// other IPC knob and armed only where the server publishes kCapRunAheadApply.
@@ -412,6 +451,10 @@ namespace MobileGL::MG_ConfigLoader {
// deeper queue buys nothing on a CPU-bound client and pays for it in latency. 0 is NOT
// admitted: a credit of zero would mean "publish no present at all".
ipc.PresentCredit = QueryEnvUint32("MOBILEGL_IPC_PRESENT_CREDIT", 1, 1, 8);
// CONTRACT-P6 D5b's reply bound and P7's cold-start budget (Config.h has the semantics).
// Both are floored at 100 ms: a zero bound would declare every live server silent.
ipc.ControlTimeoutMs = QueryEnvUint32("MOBILEGL_IPC_CONTROL_TIMEOUT_MS", 5000, 100, 600000);
ipc.ColdStartMs = QueryEnvUint32("MOBILEGL_IPC_COLD_START_MS", 20000, 100, 600000);
ipc.StrictErrors = QueryEnvFlag("MOBILEGL_IPC_STRICT_ERRORS");
ipc.Audit = QueryEnvFlag("MOBILEGL_IPC_AUDIT");
QueryEnvVariable("MOBILEGL_IPC_SERVER_AFFINITY", ipc.ServerAffinity, "auto");
@@ -451,21 +494,69 @@ namespace MobileGL::MG_ConfigLoader {
}
}
// P12 (on-screen): MOBILEGL_IPC_SURFACE = offscreen | server | host (Config.h has the
// semantics). An unknown value is named and read as the default, like every knob here.
{
String surface;
QueryEnvVariable("MOBILEGL_IPC_SURFACE", surface, "offscreen");
std::transform(surface.begin(), surface.end(), surface.begin(),
[](unsigned char c) { return static_cast<char>(std::tolower(c)); });
if (surface.empty() || surface == "offscreen") {
ipc.Surface = MG_Config::IpcSurface::Offscreen;
} else if (surface == "server") {
ipc.Surface = MG_Config::IpcSurface::Server;
} else if (surface == "host") {
ipc.Surface = MG_Config::IpcSurface::Host;
} else {
MGLOG_W("Config: Ignoring invalid env variable MOBILEGL_IPC_SURFACE='%s'; expected "
"offscreen|server|host, using offscreen",
surface.c_str());
ipc.Surface = MG_Config::IpcSurface::Offscreen;
}
// (Not said by a server process: a spawned server inherits the client's environment,
// parses this before RunSession sets its own transport, and has no client half anyway.)
const char* role = std::getenv("MOBILEGL_IPC_ROLE");
const Bool serverProcess = role != nullptr && std::strcmp(role, "server") == 0;
if (ipc.Surface != MG_Config::IpcSurface::Offscreen && !serverProcess &&
MG_Config::Transport != MG_Config::TransportMode::Spawn) {
// Said, not silently honoured: there is no remote server here to own a window.
MGLOG_W("Config: MOBILEGL_IPC_SURFACE=%s is IGNORED - it asks for an on-screen "
"surface (the server's own window, or the frames the display host owns) and "
"this run has no remote server to provide one: MOBILEGL_TRANSPORT is not "
"spawn. Window surfaces stay the client's own",
ipc.Surface == MG_Config::IpcSurface::Server ? "server" : "host");
}
}
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
// One line, on the arm where these numbers decide behaviour, because every one of
// them is a number a bug report has to quote.
MGLOG_I("Config: IPC ring=%uMiB stage=%uMiB spin=%uus persistent-block=%uKiB "
"adopt-tier=%u verb-barrier=%u run-ahead=%u present-credit=%u strict=%d "
"audit=%d role-split-state=%d affinity='%s'",
ipc.RingMb, ipc.StageMb, ipc.SpinUs, ipc.PersistentBlockKb, ipc.AdoptTier,
ipc.VerbBarrier, ipc.RunAhead, ipc.PresentCredit,
// CREATE-WINDOW IS IN THIS LINE BECAUSE A RUN HAS TO SAY WHICH WINDOW IT USED. It was
// missing while the value was a number nobody varied, and the moment it became one that
// changes the wall clock by seconds (the P12 sweep: 28.3 s at 1 down to 10.8 s at 6) a
// log that does not name it makes every run ambiguous about its own configuration.
MGLOG_I("Config: IPC ring=%uMiB stage=%uMiB wire-deferred=%uMiB spin=%uus event-wait=%ums "
"persistent-block=%uKiB create-window=%u pbo-readback-sync=%u poll-escalate=%u "
"adopt-tier=%s allow-t0=%u verb-barrier=%u run-ahead=%u present-credit=%u control-timeout=%ums "
"cold-start=%ums strict=%d audit=%d role-split-state=%d affinity='%s' surface=%s",
ipc.RingMb, ipc.StageMb, ipc.WireDeferredMb, ipc.SpinUs, ipc.EventWaitMs,
ipc.PersistentBlockKb, ipc.CreateWindow, ipc.PboReadbackSync, ipc.PollEscalate,
ipc.AdoptTier == 0 ? "0" : ipc.AdoptTier == 1 ? "1" : ipc.AdoptTier == 2 ? "2" : "unset(T0)",
ipc.AllowAdoptT0,
ipc.VerbBarrier, ipc.RunAhead, ipc.PresentCredit, ipc.ControlTimeoutMs, ipc.ColdStartMs,
static_cast<int>(ipc.StrictErrors), static_cast<int>(ipc.Audit),
static_cast<int>(ipc.RoleSplitState), ipc.ServerAffinity.c_str());
static_cast<int>(ipc.RoleSplitState), ipc.ServerAffinity.c_str(),
MG_Config::ServerOwnedWindowSurfaces() ? "server" : "offscreen");
if (ipc.VerbBarrier == 0) {
MGLOG_W("Config: MOBILEGL_IPC_VERB_BARRIER=0 is the R-1 NEGATIVE CONTROL and is "
"expected to fail: the client still pulls 31 of 63 PipeInputs fields from a "
"live GLContext, so an unbarriered queue lets the server read future values");
}
if (ipc.WireDeferredMb == 0) {
MGLOG_W("Config: MOBILEGL_IPC_WIRE_DEFERRED_MB=0 is the M2 NEGATIVE CONTROL: the "
"server never forces a sync for orphaned wire buffer stores, so a long frame "
"that respecifies and draws holds every one of them until the frame ends");
}
if (ipc.PersistentBlockKb == 0) {
MGLOG_W("Config: MOBILEGL_IPC_PERSISTENT_BLOCK_KB=0 is the E3(a) NEGATIVE CONTROL: "
"the persistent-map push is OFF and a coherent-map scenario must go red");
+9 -2
View File
@@ -8,6 +8,8 @@
#pragma once
#include <cstdio>
// ============== Platform-specific definitions and macros ============== //
// No __ANDROID_API__ pin here on purpose. The effective API level is owned by
// the build system (gradle minSdk 26 -> -DANDROID_PLATFORM=android-26, enforced
@@ -34,6 +36,11 @@
#ifdef _WIN32
#define MOBILEGL_EXPORT extern "C" __declspec(dllexport)
#else
// Default visibility, which is what the platform's own GL and EGL headers declare their entry
// points with. The one place where that is not enough - a vendor library handing glvnd its own
// entry points, where a reference resolving against the process instead of this library would be a
// call back into the dispatcher - is answered at link time with -Bsymbolic-functions rather than
// here, because a second visibility on the same declaration does not compile.
#define MOBILEGL_EXPORT extern "C" __attribute__((visibility("default")))
#endif
@@ -105,8 +112,8 @@
#define MOBILEGL_ASSERT(condition, ...) \
do { \
if (!(condition)) { \
MGLOG_F("Assertion failed" __VA_OPT__(": ") __VA_ARGS__); \
MGLOG_F(" at %s:%d (%s)", __FILE__, __LINE__, __func__); \
fprintf(stderr, "Assertion failed" __VA_OPT__(": ") "%s", "" __VA_ARGS__); \
fprintf(stderr, " at %s:%d (%s)\n", __FILE__, __LINE__, __func__); \
TRAP; \
} \
} while (0)
+53 -1
View File
@@ -21,6 +21,9 @@
#include <MG_Util/ShaderTranspiler/ShaderCompiler.h>
#include <MG_State/GLState/ProgramState/ProgramTranslationCache.h>
#include <MG_Util/ShaderTranspiler/TranslationCache.h>
#if MOBILEGL_PIPE_VERIFY
#include <MG_Backend/MGPipe/PipeInputs.h> // MGPipeVerifyFlushSummary, before Debug::Close in DestroyImpl
#endif
#include <atomic>
#include <mutex>
@@ -99,6 +102,14 @@ namespace MobileGL {
MG_State::GLState::ClearProgramTranslationCache();
MG_Backend::gBackendFunctionsTable = {};
g_isInitialized = false;
#if MOBILEGL_PIPE_VERIFY
// BEFORE Close, and from here rather than from a static destructor (V1 fix round 2):
// Close() nulls the role's sink and Log.cpp's next write reopens it with "w", so the
// FATAL=0 summary the comparator used to write from __run_exit_handlers truncated the
// client half of every such run to that one line. Verify builds only - the guard is
// the compile definition, not a runtime knob, and the pull build gains no byte (G1).
MG_Pipe::MGPipeVerifyFlushSummary();
#endif
if (logLifecycle) {
MG_Util::Debug::Close();
}
@@ -121,9 +132,50 @@ namespace MobileGL {
// boundary counters latch their enable flag here, so every counting site in the
// two backends is a load of an already-settled global for the rest of the run.
MG_Util::PipeStats::Init();
#if MOBILEGL_BUILD_DISAGGREGATED
// F1 (P7 wave 2). THE WIRE COMES UP BEFORE THE FIRST FRONTEND OBJECT IS BORN, and the
// ORDER is the fix - not a wait, and not a friendlier default mask.
//
// MG_State::Init() constructs GLContext, and GLContext's TextureState constructor
// materialises one default texture object per target (GL 3.3 core 3.8,
// TextureState.cpp:62-71). TextureObjectBase's constructor emits resource_create for
// every one of them (TextureState/TextureObject.cpp:168-169), and that emission is
// gated, through PipeFill's FamilyIsLive, on
// CapsMirror::ServerConsumes(kMGPipeSubsystemTextureResources). In the old order the
// mirror was still GENERATION 0 when it was asked - it had no snapshot because the
// session had not been started yet, MG_Backend::Init() being what starts it - so it
// answered from the placeholder, every one of those creates was withheld, and the log
// said so once and then never again:
//
// MG_Remote client: the server does not consume MGPipe subsystem 0x400 - this
// family emits NOTHING and the legacy pull path runs for it (R-8). callMask=0x0,
// caps generation 0
//
// That is not a race and it never was one: the line reproduces byte-for-byte on
// lavapipe and on an Adreno 830, because the two Init calls are two statements in one
// thread. And "the legacy pull path runs for it" is false under split - the client has
// no pull path to fall back to - so the records are simply lost, and only a later
// respecify's self-healing create (TextureEmit.h:872) ever brings one back.
//
// A BOUNDED WAIT INSIDE ServerConsumes CANNOT FIX THIS AND WOULD DEADLOCK: during
// MG_State::Init() no handshake is in flight, and the one that will land is started
// LATER BY THIS SAME THREAD. The only correct answer is to have already brought it up,
// which is what this hook does. ServerConsumes now REFUSES to answer from a
// placeholder (CapsMirror.cpp), so a regression of this order is a named Fatal rather
// than a whole family of silently dropped records.
//
// In a build without MOBILEGL_BUILD_DISAGGREGATED this statement does not exist, so
// the pull build gains no symbol, no branch and no byte - which is what G1 measures.
const Bool splitBackendIsUp = MG_Backend::InitSplitRolesBeforeState();
#endif
MG_State::Init();
MGLOG_D("MG_State initialized");
MG_Backend::Init();
#if MOBILEGL_BUILD_DISAGGREGATED
if (!splitBackendIsUp)
#endif
{
MG_Backend::Init();
}
MGLOG_D("MG_Backend initialized");
MG_Impl::Init();
MGLOG_D("MG_Impl initialized");
+86 -5
View File
@@ -172,11 +172,15 @@ namespace MobileGL::MG_Backend {
return false;
}
if (m_eglDisplayInitialized && m_eglDisplay != dpy) {
MGLOG_E("InitializeEGLDisplay failed: backend already bound to a different EGLDisplay");
return false;
}
// A DISPLAY IS A CONNECTION, NOT A RESOURCE THIS BACKEND OWNS, and refusing the second one
// refused a caller for a reason that is not there: the surface and context tables are keyed
// by the display each was created on, every operation carries its own display handle, and
// this field has no reader anywhere - it was the whole of the rule.
//
// What the rule cost was a compositor. One asks for the display it will draw into AFTER
// probing another, so the second eglInitialize is the one that matters, and a refusal there
// is reported one layer up as "no usable DRM render device" - the probe having succeeded
// and the real display having failed, with nothing in between that says which was refused.
m_eglDisplay = dpy;
m_eglDisplayInitialized = true;
if (major) {
@@ -209,6 +213,28 @@ namespace MobileGL::MG_Backend {
return true;
}
// The frame itself is the backend's business (InitHostFrameSurface); what happens here is
// the registration every surface kind shares, so the swap path and the state mirrors see
// a host-framed surface exactly like any other.
Bool BackendObject::CreateEGLHostFrameSurface(EGLSurface surface, Uint32 width, Uint32 height) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (!m_eglDisplayInitialized) {
MGLOG_E("CreateEGLHostFrameSurface failed: EGL display is not initialized");
return false;
}
if (surface == EGL_NO_SURFACE) {
MGLOG_E("CreateEGLHostFrameSurface failed: invalid EGLSurface");
return false;
}
auto& state = m_eglSurfaces[surface];
state = EGLSurfaceState{
.Kind = SurfaceKind::HostFrame,
.Width = static_cast<EGLint>(std::max<Uint32>(width, 1)),
.Height = static_cast<EGLint>(std::max<Uint32>(height, 1)),
};
return ActivateEGLSurface(surface);
}
Bool BackendObject::CreateEGLPbufferSurface(EGLSurface surface, EGLint width, EGLint height) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
return RegisterEGLPbufferSurface(surface, width, height) && ActivateEGLSurface(surface);
@@ -290,6 +316,11 @@ namespace MobileGL::MG_Backend {
MGLOG_E("ActivateEGLSurface failed: backend InitPbufferSurface failed");
return false;
}
} else if (surfaceState->Kind == SurfaceKind::HostFrame) {
if (!InitHostFrameSurface(surfaceState->Width, surfaceState->Height)) {
MGLOG_E("ActivateEGLSurface failed: backend InitHostFrameSurface failed");
return false;
}
} else {
MGLOG_E("ActivateEGLSurface failed: unsupported surface kind");
return false;
@@ -315,6 +346,29 @@ namespace MobileGL::MG_Backend {
MGLOG_E("MakeEGLCurrent failed: EGL display mismatch or not initialized");
return false;
}
// EGL_KHR_surfaceless_context. A context whose draw and read surfaces are both absent has
// no framebuffer of its own to be bound to: the caller draws into framebuffers it binds
// itself, and this side only has to have its own GL context current on this thread for the
// records that follow. Everything below this point describes a surface, so none of it
// applies - and what is recorded is exactly what was asked for, so a later swap on a
// surface that does not exist is never made to look like one that does.
if (draw == EGL_NO_SURFACE && read == EGL_NO_SURFACE && ctx != EGL_NO_CONTEXT) {
if (!m_backendCapabilitiesInitialized) {
if (!InitCapabilities()) {
MGLOG_E("MakeEGLCurrent failed: InitCapabilities failed");
return false;
}
m_backendCapabilitiesInitialized = true;
}
ReleaseEGLCurrentThread(threadKey);
m_eglCurrentThreads[threadKey] = EGLCurrentState{
.Display = dpy,
.DrawSurface = EGL_NO_SURFACE,
.ReadSurface = EGL_NO_SURFACE,
.Context = ctx,
};
return true;
}
if (!m_eglSurfaceInitialized) {
if (draw != read || !ActivateEGLSurface(draw)) {
MGLOG_E("MakeEGLCurrent failed: EGL surface is not initialized");
@@ -387,6 +441,12 @@ namespace MobileGL::MG_Backend {
return false;
}
// A host-framed surface has no swapchain: presenting it means telling the host the frame
// is drawn, because the host is what puts it on the glass.
if (m_eglSurfaceKind == SurfaceKind::HostFrame) {
return PresentHostFrame();
}
const auto& backendFunctions = GetBackendFunctions();
if (!backendFunctions.Present) {
MGLOG_E("SwapEGLBuffers failed: backend Present function is null");
@@ -404,6 +464,27 @@ namespace MobileGL::MG_Backend {
}
}
// A backend that has no host-framed surfaces says so by name: the server may ask for one
// (MOBILEGL_IPC_SURFACE=host) against a build that cannot serve it, and that has to be a
// refusal with a reason rather than a surface that draws nowhere.
Bool BackendObject::InitHostFrameSurface(EGLint width, EGLint height) {
(void)width;
(void)height;
MGLOG_E("InitHostFrameSurface: this backend has no host-framed surfaces");
return false;
}
Bool BackendObject::PresentHostFrame() {
MGLOG_E("PresentHostFrame: this backend has no host-framed surfaces");
return false;
}
Bool BackendObject::PresentHostFrameIfCurrent() {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
if (!m_eglSurfaceInitialized || m_eglSurfaceKind != SurfaceKind::HostFrame) return true;
return PresentHostFrame();
}
Bool BackendObject::IsEGLSurfaceCurrent(EGLSurface surface) const {
if (surface == EGL_NO_SURFACE) {
return false;
+22 -6
View File
@@ -332,13 +332,13 @@ namespace MobileGL {
};
struct DynamicBackendParameters {
SizeT UniformBufferOffsetAlignment = 256;
Uint64 UniformBufferOffsetAlignment = 256;
// GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT, which is a SEPARATE limit from the
// uniform one and is routinely larger: Adreno 830 reports 32 for uniform buffers and
// 64 for storage buffers. Answering the storage query with the uniform value let an
// application bind a storage range at an offset the driver cannot address, which it
// accepted without error and then wrote somewhere else entirely.
SizeT ShaderStorageBufferOffsetAlignment = 256;
Uint64 ShaderStorageBufferOffsetAlignment = 256;
// GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT. 1.0 means the backend cannot filter anisotropically,
// which is also why the extension is not advertised in that case.
Float MaxTextureMaxAnisotropy = 1.0f;
@@ -460,8 +460,8 @@ namespace MobileGL {
// VulkanRenderer::SelectProvokingVertexMode out of VK_EXT_provoking_vertex,
// provokingVertexModePerPipeline and the topology, so no single convention is true
// of the backend.
GLenum LayerProvokingVertex = GL_UNDEFINED_VERTEX;
GLenum ViewportIndexProvokingVertex = GL_UNDEFINED_VERTEX;
Uint32 LayerProvokingVertex = GL_UNDEFINED_VERTEX;
Uint32 ViewportIndexProvokingVertex = GL_UNDEFINED_VERTEX;
Int MaxViewportWidth = 16384;
Int MaxViewportHeight = 16384;
Float ViewportBoundsRangeMin = 0.0f;
@@ -558,7 +558,7 @@ namespace MobileGL {
Bool SupportsTessellationPointSize = true;
// The geometry-stage twin (OpCapability GeometryPointSize).
Bool SupportsGeometryPointSize = true;
SizeT MaxShaderStorageBlockSize = 128 * 1024 * 1024;
Uint64 MaxShaderStorageBlockSize = 128 * 1024 * 1024;
Uint32 SubgroupSize = 0;
Uint32 SubgroupSupportedStages = 0;
Uint32 SubgroupSupportedFeatures = 0;
@@ -595,6 +595,18 @@ namespace MobileGL {
virtual Bool CreateEGLWindowSurface(EGLSurface surface, const WindowHandle& handle);
virtual Bool ResizeEGLWindowSurface(EGLSurface surface, Uint32 width, Uint32 height);
virtual Bool CreateEGLPbufferSurface(EGLSurface surface, EGLint width, EGLint height);
// The host owns the screen: this side never gets a window of the client's to draw
// into, so the surface it draws into is a frame the host allocated and offered over
// the host frame bridge. Each swap tells the host the frame is done and takes the
// next; presentation is the host's, not a swapchain's.
virtual Bool CreateEGLHostFrameSurface(EGLSurface surface, Uint32 width, Uint32 height);
virtual Bool InitHostFrameSurface(EGLint width, EGLint height);
virtual Bool PresentHostFrame();
// The present RECORD's half of a host-framed swap. A split client's swap reaches the
// server as a present record (ServerVerbSink::OnPresent), never as SwapEGLBuffers, so
// without this the host was never told a frame was drawn. True when the current
// surface is not host-framed: there is nothing to present then.
Bool PresentHostFrameIfCurrent();
virtual Bool MakeEGLCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext ctx);
virtual Bool SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw);
// Forwards the app-requested eglSwapInterval to the backend's native
@@ -616,7 +628,11 @@ namespace MobileGL {
enum class SurfaceKind {
None,
Window,
Pbuffer
Pbuffer,
// A surface whose drawing target is a frame the display host owns. There is no
// swapchain behind it: the host puts the frame on the glass when this side says
// it is drawn, which is why presenting one is its own call and not a Present.
HostFrame
};
struct EGLCurrentState {
+16
View File
@@ -39,5 +39,21 @@ namespace MobileGL::MG_Backend {
// client-minted and needs no backend call, which is P10's. CONTRACT-P5 4 flags this as a
// KNOWN OPEN ITEM and asks v1 to record which way it went: P5 keeps today's order.
void ShutdownSplitRoles();
// F1 (P7 wave 2). Init()'s SPLIT ARM, run BEFORE MG_State::Init() rather than after it.
//
// Returns true when this process took the split path - whether the bring-up succeeded or
// was refused by name - and the caller must then NOT call Init() a second time. Returns
// false under MG_Config::TransportMode::Monolith, where there is nothing to bring up and
// Init() keeps its own place after MG_State::Init().
//
// WHY IT EXISTS AT ALL, rather than MobileGL::Initialize simply calling Init() earlier:
// the monolith arm of Init() must keep running after MG_State::Init(), because
// BackendObject_DirectGLES::Initialize() is the tree's own order for it and moving it is a
// pull-build change G1 forbids. So the split arm - and only the split arm - moves.
//
// It exists ONLY in a split build, for the reason InitServerRoleForSpawn states: a
// definition guarded only on the inside still emits a symbol, and G1 measured that.
Bool InitSplitRolesBeforeState();
#endif
} // namespace MobileGL::MG_Backend
@@ -6,6 +6,12 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#if MOBILEGL_BUILD_DISAGGREGATED && defined(__ANDROID__)
// The role's own backend is resolved the way Utils.cpp (C6 / ID-52) resolves it; that needs this
// header, and it has to be included at file scope - a header included inside this file's namespace
// would nest its own namespace under it.
#include <MG_Remote/Server/ServerLoop.h>
#endif
#include "BackendObject_DirectGLES.h"
#include "MG_Backend/BackendObject.h"
#include "MG_Backend/BackendObjects.h"
@@ -893,6 +899,16 @@ namespace MobileGL::MG_Backend::DirectGLES {
DestroyEGLContext();
#if MOBILEGL_BUILD_DISAGGREGATED
NoteNativeContextGone();
// P12: under a transport this object is the SERVER's backend and its destruction is the
// end of a session (ServerLoop's apply thread, after the final drain). The in-process
// display server runs the next session in this same process, so the twins this one built
// - naming ids of the context just destroyed - must not answer for the next client's
// handles (Managers.h, DropEveryTwinForEndedServerSession). Monolith keeps its twins.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
DropEveryTwinForEndedServerSession();
// Review fix: and the swap interval it asked for (DirectGLES.h).
ForgetRequestedSwapInterval();
}
#endif
}
@@ -1018,6 +1034,241 @@ namespace MobileGL::MG_Backend::DirectGLES {
#endif
}
#if MOBILEGL_BUILD_DISAGGREGATED && defined(__ANDROID__)
namespace {
// The host may still be starting when the client asks for its surface: it is a separate
// app that has to come up and allocate its pool. Waiting is better than refusing a
// surface a client is entitled to, and the wait is answered by name if it expires.
constexpr Int kHostFrameWaitMs = 30000;
}
Bool TakeHostFrameForExternalTexture(Uint width, Uint height) {
// THIS ROLE'S OWN BACKEND, resolved exactly the way ActiveBackendFormatCaps (Utils.cpp,
// C6 / ID-52) resolves it: under a live split session the SERVER's private backend owns the
// context on the apply thread, and pActiveBackendObject is the CLIENT's mirror there.
MG_Backend::BackendObject* backend = nullptr;
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
backend = MG_Remote::Server::ServerLoopInstance().Backend();
} else {
backend = MG_Backend::pActiveBackendObject.get();
}
if (backend == nullptr) return false;
return backend->InitHostFrameSurface(static_cast<EGLint>(width), static_cast<EGLint>(height));
}
Bool BackendObject_DirectGLES::InitHostFrameSurface(EGLint width, EGLint height) {
// The EGL context comes from the same place a pbuffer session's does. What changes is not
// how this side draws but what it draws INTO: a frame the display host allocated, bound as
// this context's framebuffer. The pbuffer stays as the draw target for the moment before
// the first frame arrives, which is the only moment it is used.
if (!DirectGLES::InitPbufferSurface(width, height)) {
MGLOG_E("InitHostFrameSurface: the context the host frames are drawn in did not come up");
return false;
}
if (!m_hostFrameBridgeOpen) {
String why;
if (!m_hostFrameBridge.Open(MG_Backend::kHostFrameSocket, kHostFrameWaitMs, why)) {
MGLOG_E("InitHostFrameSurface: no display host: %s", why.c_str());
return false;
}
m_hostFrameBridgeOpen = true;
}
return TakeHostFrame();
}
Bool BackendObject_DirectGLES::TakeHostFrame() {
String why;
struct AHardwareBuffer* frame = nullptr;
MG_Backend::HostFrameOffer offer{};
if (!m_hostFrameBridge.Acquire(&frame, offer, why)) {
MGLOG_E("TakeHostFrame: %s", why.c_str());
return false;
}
HostFrameTarget target;
if (!HostFrameTargetCreate(frame, target, why)) {
MGLOG_E("TakeHostFrame: the host frame %ux%u did not become a draw target: %s", offer.Width,
offer.Height, why.c_str());
AHardwareBuffer_release(frame);
return false;
}
DestroyHostFrame();
m_hostFrameTarget = target;
m_hostFrameBuffer = frame;
m_hostFrameOffer = offer;
// The canvas is made at the first frame's size and kept; a failure leaves the frame itself
// as the draw target, which is how this path behaved before the canvas existed.
if (!EnsureHostFrameCanvas(target.Width, target.Height))
MGLOG_E_ONCE("TakeHostFrame: no canvas in front of the host frames; drawing into the frame itself");
const HostFrameTarget& drawTarget = HostFrameDrawTarget();
// BEFORE the bind: every default-framebuffer bind the engine makes from here on has to
// resolve to this target, and that resolution is what this global is.
g_hostFrameFramebufferId = drawTarget.Framebuffer;
// Published together with the framebuffer, because they describe the same target: the STORAGE
// an attachment is bound to when the client's texture has none of its own, and the SIZE that
// decides whether binding it is right at all.
g_hostFrameRenderbuffer = drawTarget.Renderbuffer;
g_hostFrameWidth = drawTarget.Width;
g_hostFrameHeight = drawTarget.Height;
FramebufferImpl::InvalidateFramebufferBindingCache();
HostFrameTargetBind(drawTarget);
// AND ON TO THE CONTAINER, as the dma-buf descriptor behind it. The Wayland compositor
// over there renders into THESE frames - it runs this project's own GBM, whose
// gbm_bo_create answers with the phone's frame instead of an allocation of its own - so
// every frame taken here is useless to it until its descriptor has crossed. Open() binds
// the listening socket a container dials, and the IsUp() test keeps the ordinary frame
// down to an integer compare: the socket outlives every container and every connection,
// while OfferFrame accepts whoever dialled since the last frame and opens the socket
// itself when the first frame arrives before any container has.
//
// NEITHER ANSWER IS EXAMINED, and that is the channel's contract rather than an
// oversight: no container running, a container that has died, or one that has taken the
// descriptor and stopped reading must all leave this frame drawn and presented exactly as
// it would have been without a channel - the descriptor is a copy of the pixels' route,
// not a step in producing them.
if (!m_gbmFrameChannel.IsUp()) m_gbmFrameChannel.Open();
(void)m_gbmFrameChannel.OfferFrame(frame, offer);
MGLOG_I("host frame %u is the draw target: %ux%u", offer.Index, target.Width, target.Height);
return true;
}
void BackendObject_DirectGLES::DestroyHostFrame() {
// The canvas outlives the frame, so with one the globals keep naming it: the attachments
// made over it stay valid between this frame being handed back and the next one arriving.
g_hostFrameFramebufferId = m_hostFrameCanvas.Framebuffer;
g_hostFrameRenderbuffer = m_hostFrameCanvas.Renderbuffer;
g_hostFrameWidth = m_hostFrameCanvas.Width;
g_hostFrameHeight = m_hostFrameCanvas.Height;
HostFrameTargetDestroy(m_hostFrameTarget);
if (m_hostFrameBuffer != nullptr) {
AHardwareBuffer_release(m_hostFrameBuffer);
m_hostFrameBuffer = nullptr;
}
// AND NOTHING IS SAID TO THE CHANNEL HERE, which is the fix rather than an omission. This
// runs once per presented frame: it used to Close() the channel (unlinking the container's
// socket node and dropping its connection sixty times a second - measured: connect()
// answered EAGAIN(11) on every attempt) and it used to tell the channel to forget the frame,
// which left the ONE window a compositor's first allocations arrive in - between this frame
// being presented and the host handing the next one over, i.e. inside Complete() above -
// with nothing to answer a dialer with. A compositor allocates its swapchain ONCE and keeps
// those buffers for the life of the process, so a dialer that finds nothing here renders
// its whole life into buffers nobody displays. The frame the channel remembers is the last
// one the host handed over: the next offer replaces it and Close() releases it.
}
Bool BackendObject_DirectGLES::PresentHostFrame() {
// When this returns, the pixels are complete: glFinish is the fence the host is told
// about, and the host's own readback - which Complete reads - is what proves the drawing
// landed where the screen will look.
BlitHostFrameCanvas();
if (g_GLESFuncs.glFinish != nullptr) g_GLESFuncs.glFinish();
// WHAT THE CONTAINER SAID WHILE THIS FRAME WAS BEING DRAWN. This is the one place the
// server's apply loop passes through the host channel - the swap it runs lands here for a
// host-framed surface - and a release is the container's compositor saying it is done
// reading a frame this side handed over. Poll() is a bare recvmsg on a non-blocking
// socket: with no container connected it returns on its first test, and a container that
// is between frames answers EAGAIN, so a pass costs nothing when there is nothing to
// read. Nothing here is waited for and nothing is required to arrive; the count of
// frames offered and not released is a number the channel keeps for the log, never a
// frame the render path waits on.
m_gbmFrameChannel.Poll();
String why;
if (!m_hostFrameBridge.Complete(m_hostFrameOffer, /*imported=*/true, /*drawn=*/true, /*fenceOk=*/true, why)) {
MGLOG_E("PresentHostFrame: %s", why.c_str());
return false;
}
// The answer has been read, so the host has already put this frame on the glass: it is
// this side's to drop, and the next one becomes the drawing target.
DestroyHostFrame();
return TakeHostFrame();
}
Bool BackendObject_DirectGLES::EnsureHostFrameCanvas(Uint width, Uint height) {
if (m_hostFrameCanvas.Framebuffer != 0) return true;
if (width == 0 || height == 0 || !g_GLESFuncs.glRenderbufferStorage || !g_GLESFuncs.glBlitFramebuffer ||
!g_GLESFuncs.glGenRenderbuffers || !g_GLESFuncs.glGenFramebuffers || !g_GLESFuncs.glCheckFramebufferStatus)
return false;
// RAW binds, as HostFrameTargetCreate makes them: the caller invalidates the engine's
// framebuffer shadow and binds through it right after.
GLuint renderbuffer = 0;
g_GLESFuncs.glGenRenderbuffers(1, &renderbuffer);
g_GLESFuncs.glBindRenderbuffer(GL_RENDERBUFFER, renderbuffer);
// The frames are R8G8B8A8_UNORM (the display host allocates them so), and so is this.
g_GLESFuncs.glRenderbufferStorage(GL_RENDERBUFFER, GL_RGBA8, static_cast<GLsizei>(width),
static_cast<GLsizei>(height));
GLuint framebuffer = 0;
g_GLESFuncs.glGenFramebuffers(1, &framebuffer);
g_GLESFuncs.glBindFramebuffer(GL_FRAMEBUFFER, framebuffer);
g_GLESFuncs.glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, renderbuffer);
const GLenum status = g_GLESFuncs.glCheckFramebufferStatus(GL_FRAMEBUFFER);
if (status != GL_FRAMEBUFFER_COMPLETE) {
MGLOG_E("host frame canvas %ux%u: framebuffer status 0x%04x", width, height, status);
g_GLESFuncs.glDeleteFramebuffers(1, &framebuffer);
g_GLESFuncs.glDeleteRenderbuffers(1, &renderbuffer);
return false;
}
m_hostFrameCanvas.Renderbuffer = renderbuffer;
m_hostFrameCanvas.Framebuffer = framebuffer;
m_hostFrameCanvas.Width = width;
m_hostFrameCanvas.Height = height;
MGLOG_I("host frame canvas: %ux%u as framebuffer %u (renderbuffer %u), copied into each frame at present",
width, height, framebuffer, renderbuffer);
return true;
}
void BackendObject_DirectGLES::BlitHostFrameCanvas() {
if (m_hostFrameCanvas.Framebuffer == 0 || m_hostFrameTarget.Framebuffer == 0) return;
// Whatever the compositor had bound is put back as it was, raw for raw, so the engine's
// shadow (which never saw this) is still right afterwards. The scissor test and
// rasterizer discard both apply to a blit, so they are lifted for it.
GLint readBinding = 0;
GLint drawBinding = 0;
g_GLESFuncs.glGetIntegerv(GL_READ_FRAMEBUFFER_BINDING, &readBinding);
g_GLESFuncs.glGetIntegerv(GL_DRAW_FRAMEBUFFER_BINDING, &drawBinding);
const GLboolean scissor = g_GLESFuncs.glIsEnabled(GL_SCISSOR_TEST);
const GLboolean discard = g_GLESFuncs.glIsEnabled(GL_RASTERIZER_DISCARD);
if (scissor) g_GLESFuncs.glDisable(GL_SCISSOR_TEST);
if (discard) g_GLESFuncs.glDisable(GL_RASTERIZER_DISCARD);
g_GLESFuncs.glBindFramebuffer(GL_READ_FRAMEBUFFER, m_hostFrameCanvas.Framebuffer);
g_GLESFuncs.glBindFramebuffer(GL_DRAW_FRAMEBUFFER, m_hostFrameTarget.Framebuffer);
const Bool sameSize = m_hostFrameCanvas.Width == m_hostFrameTarget.Width &&
m_hostFrameCanvas.Height == m_hostFrameTarget.Height;
g_GLESFuncs.glBlitFramebuffer(0, 0, static_cast<GLint>(m_hostFrameCanvas.Width),
static_cast<GLint>(m_hostFrameCanvas.Height), 0, 0,
static_cast<GLint>(m_hostFrameTarget.Width),
static_cast<GLint>(m_hostFrameTarget.Height), GL_COLOR_BUFFER_BIT,
sameSize ? GL_NEAREST : GL_LINEAR);
g_GLESFuncs.glBindFramebuffer(GL_READ_FRAMEBUFFER, static_cast<GLuint>(readBinding));
g_GLESFuncs.glBindFramebuffer(GL_DRAW_FRAMEBUFFER, static_cast<GLuint>(drawBinding));
if (scissor) g_GLESFuncs.glEnable(GL_SCISSOR_TEST);
if (discard) g_GLESFuncs.glEnable(GL_RASTERIZER_DISCARD);
}
#else // !(MOBILEGL_BUILD_DISAGGREGATED && __ANDROID__)
// The other arm's half: on a build with no host frames there is no frame to take, and the
// attach site's ordinary path (a storage-less external texture, exactly as before) is the honest
// behaviour there rather than a link error.
Bool TakeHostFrameForExternalTexture(Uint, Uint) { return false; }
Bool BackendObject_DirectGLES::InitHostFrameSurface(EGLint width, EGLint height) {
(void)width;
(void)height;
MGLOG_E("InitHostFrameSurface: host frames are a disaggregated Android path");
return false;
}
Bool BackendObject_DirectGLES::TakeHostFrame() { return false; }
void BackendObject_DirectGLES::DestroyHostFrame() {}
Bool BackendObject_DirectGLES::EnsureHostFrameCanvas(Uint, Uint) { return false; }
void BackendObject_DirectGLES::BlitHostFrameCanvas() {}
Bool BackendObject_DirectGLES::PresentHostFrame() {
MGLOG_E("PresentHostFrame: host frames are a disaggregated Android path");
return false;
}
#endif
Bool BackendObject_DirectGLES::InitPbufferSurface(EGLint width, EGLint height) {
return DirectGLES::InitPbufferSurface(width, height);
}
@@ -1042,13 +1293,42 @@ namespace MobileGL::MG_Backend::DirectGLES {
MGLOG_E("MakeEGLCurrent failed: EGL display mismatch or not initialized");
return false;
}
if (!m_eglSurfaceInitialized) {
MGLOG_E("MakeEGLCurrent failed: EGL surface is not initialized");
return false;
// EGL_KHR_surfaceless_context, WHICH IS THE SHAPE A COMPOSITOR ASKS FOR. Its context has no
// draw surface of its own - every frame goes into a framebuffer it binds itself, and under
// this project that framebuffer is one of the display host's frames - so the surface checks
// below describe a surface that legitimately does not exist. What the server still needs is
// its own GL context current on this thread for the records that follow, and that is what
// the native bind underneath them does.
const Bool surfaceless = draw == EGL_NO_SURFACE && read == EGL_NO_SURFACE && ctx != EGL_NO_CONTEXT;
if (surfaceless && !m_eglSurfaceInitialized) {
// A DRAWABLE OF THIS SIDE'S OWN, WHICH IS WHAT A SURFACELESS CLIENT LEAVES TO US. The
// backend's native context has to be current on a surface of some kind before a single
// record of the client's can run - that is what the Init*Surface calls establish - and a
// client that makes its context current with no surface at all supplies none of its own.
// A pbuffer is the one that costs nothing to keep: it is never scanned out and never
// read, because every frame the client draws goes into a framebuffer it binds itself.
// The client is NOT told it has a surface - m_eglSurface stays EGL_NO_SURFACE - so a
// swap on one is still the refusal it was.
if (!InitPbufferSurface(1, 1)) {
MGLOG_E("MakeEGLCurrent failed: the backend's own drawable for a surfaceless context could not "
"be created");
return false;
}
m_eglSurface = EGL_NO_SURFACE;
m_eglSurfaceInitialized = true;
m_eglSurfaceKind = SurfaceKind::Pbuffer;
m_eglCurrentThreads.clear();
m_backendCapabilitiesInitialized = false;
}
if (draw == EGL_NO_SURFACE || read == EGL_NO_SURFACE || ctx == EGL_NO_CONTEXT) {
MGLOG_E("MakeEGLCurrent failed: draw/read/context is invalid");
return false;
if (!surfaceless) {
if (!m_eglSurfaceInitialized) {
MGLOG_E("MakeEGLCurrent failed: EGL surface is not initialized");
return false;
}
if (draw == EGL_NO_SURFACE || read == EGL_NO_SURFACE || ctx == EGL_NO_CONTEXT) {
MGLOG_E("MakeEGLCurrent failed: draw/read/context is invalid");
return false;
}
}
#if MOBILEGL_BUILD_DISAGGREGATED
@@ -1076,9 +1356,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
// function table instead of 1), or make NativeBindCanBeSkippedFor answer true for any tuple
// (the ID-67 control's different tuple stays at 1 native bind where 2 are required).
const Bool nativelyCurrentAlready = MG_Config::Transport != MG_Config::TransportMode::Monolith &&
m_eglSurfaceInitialized && m_eglSurface == draw &&
DirectGLES::IsBackendContextCurrentOnThisThread() &&
NativeBindCanBeSkippedFor(dpy, draw, read, ctx);
(surfaceless || (m_eglSurfaceInitialized && m_eglSurface == draw &&
NativeBindCanBeSkippedFor(dpy, draw, read, ctx)));
if (!nativelyCurrentAlready && !DirectGLES::MakeCurrent()) {
NoteNativeContextGone();
return false;
@@ -1097,6 +1377,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
#endif
return false;
}
// A HOST-FRAMED SURFACE HAS TO BE PUT BACK AFTER EVERY MAKE-CURRENT. The native call
// binds the surface's own default framebuffer - the pbuffer this path makes its context
// with - and that is not where these pixels go. Both the driver and the engine's binding
// shadow have to be told again, or the next default-framebuffer bind is deduped away and
// every draw lands in a pbuffer nobody sees.
if (m_eglSurfaceKind == SurfaceKind::HostFrame && m_hostFrameTarget.Framebuffer != 0) {
FramebufferImpl::InvalidateFramebufferBindingCache();
HostFrameTargetBind(HostFrameDrawTarget());
}
return true;
}
@@ -1253,7 +1542,40 @@ namespace MobileGL::MG_Backend::DirectGLES {
// Advertised with GL_NUM_PROGRAM_BINARY_FORMATS = 0, which the
// extension explicitly permits. It is also the only thing that
// exposes glProgramParameteri before GL 4.1.
E_GL_ARB_get_program_binary};
E_GL_ARB_get_program_binary,
// kwin_wayland 6.7.3 will not use a context it cannot drive, and it asks for these
// three names first: EglContext::checkSupported() (src/opengl/eglcontext.cpp) takes the
// DESKTOP-OpenGL branch for every context whose GL_VERSION is not an "OpenGL ES"
// string, and that branch requires GL_ARB_shader_objects, GL_ARB_fragment_shader and
// GL_ARB_vertex_shader. GL_VERSION here is "4.6.0 MobileGL ...", so with the names
// missing checkSupported() answered false, EglContext::create() returned nullptr,
// EglBackend::createContext() returned false, and kwin printed "Could not initialize
// rendering context" and quit before drawing a frame.
//
// The three are true statements about this backend now rather than a test unlock: every
// procedure they define is implemented in MG_Impl/GLImpl/Exporting/Definitions.cpp over
// the same shader and program objects the core entry points drive - the 39 of
// ARB_shader_objects, ARB_vertex_shader's three attrib-location entry points, and none
// at all for ARB_fragment_shader ("New Procedures and Functions: None" - it is the
// capability half, and the reason kwin's conjunction names it).
E_GL_ARB_shader_objects, E_GL_ARB_vertex_shader, E_GL_ARB_fragment_shader,
// NPOT textures: core since GL 2.0, and every host this backend can run on provides
// them. Full non-power-of-two support - mipmapping and every wrap mode, not ES 2.0's
// CLAMP_TO_EDGE-only exception - is required from ES 3.0 on, and the entries this list
// already advertises without any gate (compute shaders, shader storage buffers,
// indirect draws) need an ES 3.1 host. Nothing in the texture path rejects or pads a
// non-power-of-two size - storage and sampling are forwarded to that host - so the
// string describes what an application actually gets.
E_GL_ARB_texture_non_power_of_two,
// GL_OES_EGL_image: the entry points are real (MG_Impl/GLImpl/Exporting/Definitions.cpp)
// and this name is what makes them reachable - libepoxy in kwin resolves
// glEGLImageTargetTexture2DOES BY NAME and refuses the import path without it, printing
// "No provider of glEGLImageTargetTexture2DOES found. Requires one of: GL_OES_EGL_image".
// That import path is how the compositor's scene reaches the display daemon's dma-bufs.
// The render-server half of the import (fd over the GBM socket, then the server-side
// EGLImage) is the next change; until it lands an imported texture has no storage behind
// it, which is logged once at the entry point rather than left to look like a black frame.
E_GL_OES_EGL_image};
// Minecraft 26.3 checks this prerequisite before it even considers
// GL_ARB_multi_draw_indirect. ES 3.1 supplies both single-draw entry points; the loader
// folds the version and pointer checks into SupportsDrawIndirect.
@@ -8,6 +8,9 @@
#pragma once
#include <Includes.h>
#include "HostFrameTarget.h"
#include "MG_Backend/HostFrameBridge.h"
#include "MG_Backend/GbmFrameChannel.h"
#include "../BackendObject.h"
#include <MG_Util/BackendLoaders/OpenGL/Loader.h>
@@ -56,8 +59,46 @@ namespace MobileGL::MG_Backend::DirectGLES {
private:
void UpdateDynamicBackendParameters();
Bool InitPbufferSurface(EGLint width, EGLint height) override;
// A host-framed surface draws into a frame the display host owns; presenting one
// tells the host the frame is drawn, because the host is what puts it on the glass.
// NOT CreateEGLHostFrameSurface: the base class registers the surface and activates it,
// and activation is what reaches InitHostFrameSurface below. There is nothing about the
// EGL side of a host-framed surface this backend needs to do its own way.
Bool InitHostFrameSurface(EGLint width, EGLint height) override;
Bool PresentHostFrame() override;
// Takes the next frame from the host and makes it this context's framebuffer.
Bool TakeHostFrame();
void DestroyHostFrame();
// The canvas the compositor draws into, made once at the first frame's size.
Bool EnsureHostFrameCanvas(Uint width, Uint height);
// Copies the canvas into the frame about to be presented.
void BlitHostFrameCanvas();
// What draws resolve to: the canvas when there is one, the frame itself otherwise.
const HostFrameTarget& HostFrameDrawTarget() const {
return m_hostFrameCanvas.Framebuffer != 0 ? m_hostFrameCanvas : m_hostFrameTarget;
}
void OnEGLSurfaceReleased(EGLSurface surface) override;
// MOBILEGL_IPC_SURFACE=host: the frames the display host owns, one at a time, which
// is what the host offers - it hands one over and waits for the answer before it
// presents it and offers the next.
MG_Backend::HostFrameBridge m_hostFrameBridge;
// The frames the bridge just took, on to the container: the Wayland compositor over there
// renders into this same memory, and the only route to it that end has is the dma-buf
// descriptor this channel offers per frame and reads a release back from.
MG_Backend::GbmFrameChannel m_gbmFrameChannel;
HostFrameTarget m_hostFrameTarget;
// A STABLE TARGET IN FRONT OF THE FRAMES. Every present hands the frame back and takes
// a different one, and its renderbuffer is deleted with it - but a compositor attaches
// its framebuffers ONCE (kwin: one per daemon buffer, for the life of the process), so
// attachments bound to a frame's own renderbuffer named a deleted object from the second
// frame on. The attachments name this instead, and each present copies it into the
// frame. No EGLImage behind it (Image stays null).
HostFrameTarget m_hostFrameCanvas;
struct AHardwareBuffer* m_hostFrameBuffer = nullptr;
MG_Backend::HostFrameOffer m_hostFrameOffer{};
Bool m_hostFrameBridgeOpen = false;
Bool m_initialized = false;
MG_External::EGLFunctionsTable m_EGLFunctions;
MG_External::GLESFunctionsTable m_GLESFunctions;
File diff suppressed because it is too large Load Diff
+38 -5
View File
@@ -23,15 +23,21 @@ namespace MobileGL::MG_Backend::DirectGLES {
// Content uploads use scratch bindings, so draws and dispatches call this after
// texture synchronization.
void BindCurrentTextures();
#if MOBILEGL_BUILD_DISAGGREGATED
Bool ReadTextureImageWire(const MG_Pipe::MGPReadbackInfo& info, Vector<Uint8>& bytes);
// Native/server-only color level readback. Returns tightly packed, owned
// Driver-side color/depth level readback. Returns tightly packed, owned
// bytes in the requested pair; never consults a frontend texture or PACK/PBO.
// The caller resolves identity/extent from its resource record and supplies
// whether the SOURCE allocation already uses an image carrier.
// The caller supplies identity and extent, and whether the SOURCE allocation
// already uses an image carrier.
Bool ReadTextureLevelTight(GLuint texture, TextureTarget target, TextureUploadTarget uploadTarget,
TextureInternalFormat logicalFormat, GLint level, const IntVec3& logicalExtent,
Bool sourceUsesImageCarrier, GLenum format, GLenum type, Vector<Uint8>& bytes);
#if MOBILEGL_BUILD_DISAGGREGATED
Bool ReadTextureImageWire(const MG_Pipe::MGPReadbackInfo& info, Vector<Uint8>& bytes);
// Whether the NATIVE texture currently holds `level` of `uploadTarget` at `logicalExtent`
// (glGetTexLevelParameteriv). False for a level the driver never allocated - one defined after
// the texture's last sync - so a re-mint does not read what is not there. A query the driver
// refuses answers true, keeping the caller on its readback path.
Bool NativeTextureLevelHasExtent(GLuint texture, TextureTarget target, TextureUploadTarget uploadTarget,
GLint level, const IntVec3& logicalExtent);
#endif
void ClearBufferfi(GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil);
void ClearBufferfv(GLenum buffer, GLint drawbuffer, const GLfloat* value);
@@ -188,6 +194,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
// Applies (or defers until the window surface exists) the app-requested
// eglSwapInterval on the native EGL surface.
void SetSwapInterval(Int interval);
#if MOBILEGL_BUILD_DISAGGREGATED
// P12 review fix: the swap interval an ended server session asked for is not the next one's.
// The request outlives the context on purpose (a surface re-created mid-session re-applies it),
// so the in-process display server's session end forgets it: the next session starts at the
// driver's default until it asks (BackendObject_DirectGLES's destructor under a transport).
void ForgetRequestedSwapInterval();
#endif
void SetEGLFuncsTable(const MG_External::EGLFunctionsTable& eglFuncs);
void SetGLESFuncsTable(const MG_External::GLESFunctionsTable& glesFuncs);
void SetGLESCapabilities(const MG_External::GLESCapabilities& capabilities);
@@ -236,6 +249,26 @@ namespace MobileGL::MG_Backend::DirectGLES {
void InvalidateSyncedRenderState();
} // namespace RenderStateImpl
// MOBILEGL_IPC_SURFACE=host: what the client's DEFAULT framebuffer means while a host-framed
// surface is current. Zero everywhere else, because the driver's own 0 IS the EGL surface
// then - but on a host-framed surface the pixels belong to a frame the display host owns,
// and drawing into the driver's 0 would draw into a pbuffer nobody ever sees.
extern Uint g_hostFrameFramebufferId;
// WHAT THE HOST FRAME'S STORAGE IS, for the one caller that has to bind it in place of a
// storage-less attachment: an imported dma-buf is the display host's frame (same memory, same
// size), and the frame's storage is a RENDERBUFFER (HostFrameTarget.cpp binds it with
// glFramebufferRenderbuffer). A texture argument cannot name it, which is why these exist and
// why g_hostFrameFramebufferId alone was not enough. Zero whenever no frame is held.
extern Uint g_hostFrameRenderbuffer;
extern Uint g_hostFrameWidth;
extern Uint g_hostFrameHeight;
// TAKE THE FRAME WHEN NOTHING ELSE WILL. On the anland route the container's compositor never
// creates a window surface, so ActivateEGLSurface never runs and InitHostFrameSurface is never
// reached: the FIRST proof that the display host's frame is wanted is an ATTACHMENT of an
// imported external texture (Managers.cpp's 0x8D65 arm), and that is where the frame has to be
// taken. The host's own offer carries the width/height, so the numbers passed here are only
// what the bridge logs. Answers false (and changes nothing) whenever no backend is up.
Bool TakeHostFrameForExternalTexture(Uint width, Uint height);
extern MG_External::EGLFunctionsTable g_EGLFuncs;
extern MG_External::GLESFunctionsTable g_GLESFuncs;
extern MG_External::GLESCapabilities g_GLESCapabilities;
@@ -0,0 +1,208 @@
// MobileGL - MobileGL/MG_Backend/DirectGLES/HostFrameTarget.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "HostFrameTarget.h"
#include "DirectGLES.h"
#include "Managers.h"
#if MOBILEGL_BUILD_DISAGGREGATED && defined(__ANDROID__)
#include <android/hardware_buffer.h>
#include <cstring>
#include <format>
namespace MobileGL::MG_Backend::DirectGLES {
namespace {
// The four entry points this is spelled in are optional everywhere: the loader's own
// tables do not carry them (a missing EGL function there is MGLOG_F), so the extension
// strings decide first and the pointers are taken only once they have.
using GetNativeClientBufferFn = EGLClientBuffer (*)(const struct AHardwareBuffer*);
using CreateImageFn = EGLImageKHR (*)(EGLDisplay, EGLContext, EGLenum, EGLClientBuffer, const EGLint*);
using DestroyImageFn = EGLBoolean (*)(EGLDisplay, EGLImageKHR);
using ImageTargetRenderbufferFn = void (*)(GLenum, GLeglImageOES);
Bool HasExtensionWord(const char* list, const char* word) {
if (list == nullptr) return false;
const SizeT n = std::strlen(word);
for (const char* at = std::strstr(list, word); at != nullptr; at = std::strstr(at + 1, word)) {
const Bool startOk = at == list || at[-1] == ' ';
const Bool endOk = !at[n] || at[n] == ' ';
if (startOk && endOk) return true;
}
return false;
}
struct EntryPoints {
GetNativeClientBufferFn GetNativeClientBuffer = nullptr;
CreateImageFn CreateImage = nullptr;
DestroyImageFn DestroyImage = nullptr;
ImageTargetRenderbufferFn TargetRenderbuffer = nullptr;
};
// Resolved once per process and cached together with the refusal: a driver that lacks one
// of these does not grow it later, and the answer names what is missing.
const EntryPoints& Resolved(String& why) {
static EntryPoints points;
static Bool resolved = false;
static String refusal;
if (!resolved) {
resolved = true;
if (!g_EGLFuncs.eglGetProcAddress || !g_EGLFuncs.eglQueryString || !g_EGLFuncs.eglGetCurrentDisplay ||
!g_GLESFuncs.glGetString) {
refusal = "the loader has no eglGetProcAddress/eglQueryString/eglGetCurrentDisplay/glGetString";
} else {
const char* egl = g_EGLFuncs.eglQueryString(g_EGLFuncs.eglGetCurrentDisplay(), EGL_EXTENSIONS);
const char* gl = reinterpret_cast<const char*>(g_GLESFuncs.glGetString(GL_EXTENSIONS));
if (!HasExtensionWord(egl, "EGL_ANDROID_get_native_client_buffer")) {
refusal = "EGL_ANDROID_get_native_client_buffer is not advertised";
} else if (!HasExtensionWord(egl, "EGL_KHR_image_base")) {
refusal = "EGL_KHR_image_base is not advertised";
} else if (!HasExtensionWord(gl, "GL_OES_EGL_image")) {
refusal = "GL_OES_EGL_image is not advertised";
} else {
points.GetNativeClientBuffer = reinterpret_cast<GetNativeClientBufferFn>(
g_EGLFuncs.eglGetProcAddress("eglGetNativeClientBufferANDROID"));
points.CreateImage =
reinterpret_cast<CreateImageFn>(g_EGLFuncs.eglGetProcAddress("eglCreateImageKHR"));
points.DestroyImage =
reinterpret_cast<DestroyImageFn>(g_EGLFuncs.eglGetProcAddress("eglDestroyImageKHR"));
points.TargetRenderbuffer = reinterpret_cast<ImageTargetRenderbufferFn>(
g_EGLFuncs.eglGetProcAddress("glEGLImageTargetRenderbufferStorageOES"));
if (points.GetNativeClientBuffer == nullptr || points.CreateImage == nullptr ||
points.DestroyImage == nullptr || points.TargetRenderbuffer == nullptr) {
refusal = "eglGetProcAddress gave no eglGetNativeClientBufferANDROID / eglCreateImageKHR / "
"eglDestroyImageKHR / glEGLImageTargetRenderbufferStorageOES";
points = EntryPoints{};
}
}
}
}
why = refusal;
return points;
}
void DrainGLErrors() {
if (!g_GLESFuncs.glGetError) return;
for (Uint guard = 0; guard < 16 && g_GLESFuncs.glGetError() != GL_NO_ERROR; ++guard) {
}
}
}
Bool HostFrameTargetSupported(String& why) {
const EntryPoints& points = Resolved(why);
return points.GetNativeClientBuffer != nullptr;
}
Bool HostFrameTargetCreate(const struct AHardwareBuffer* buffer, HostFrameTarget& target, String& why) {
target = HostFrameTarget{};
if (buffer == nullptr) {
why = "no buffer";
return false;
}
const EntryPoints& points = Resolved(why);
if (points.GetNativeClientBuffer == nullptr) return false;
if (!g_GLESFuncs.glGenRenderbuffers || !g_GLESFuncs.glBindRenderbuffer || !g_GLESFuncs.glGenFramebuffers ||
!g_GLESFuncs.glBindFramebuffer || !g_GLESFuncs.glFramebufferRenderbuffer ||
!g_GLESFuncs.glCheckFramebufferStatus) {
why = "the loader has no glGenRenderbuffers/glBindRenderbuffer/glGenFramebuffers/glBindFramebuffer/"
"glFramebufferRenderbuffer/glCheckFramebufferStatus";
return false;
}
AHardwareBuffer_Desc desc{};
AHardwareBuffer_describe(buffer, &desc);
DrainGLErrors();
const EGLClientBuffer clientBuffer = points.GetNativeClientBuffer(buffer);
if (clientBuffer == nullptr) {
why = "eglGetNativeClientBufferANDROID returned null";
return false;
}
const EGLint imageAttribs[] = {EGL_IMAGE_PRESERVED_KHR, EGL_TRUE, EGL_NONE};
const EGLImageKHR image =
points.CreateImage(g_EGLFuncs.eglGetCurrentDisplay(), EGL_NO_CONTEXT, EGL_NATIVE_BUFFER_ANDROID,
clientBuffer, imageAttribs);
if (image == EGL_NO_IMAGE_KHR) {
why = "eglCreateImageKHR(EGL_NATIVE_BUFFER_ANDROID) refused the host frame";
return false;
}
GLuint renderbuffer = 0;
g_GLESFuncs.glGenRenderbuffers(1, &renderbuffer);
g_GLESFuncs.glBindRenderbuffer(GL_RENDERBUFFER, renderbuffer);
points.TargetRenderbuffer(GL_RENDERBUFFER, reinterpret_cast<GLeglImageOES>(image));
const GLenum renderbufferError = g_GLESFuncs.glGetError ? g_GLESFuncs.glGetError() : GL_NO_ERROR;
if (renderbufferError != GL_NO_ERROR) {
why = std::format("glEGLImageTargetRenderbufferStorageOES -> 0x{:04x}", renderbufferError);
g_GLESFuncs.glDeleteRenderbuffers(1, &renderbuffer);
points.DestroyImage(g_EGLFuncs.eglGetCurrentDisplay(), image);
return false;
}
GLuint framebuffer = 0;
g_GLESFuncs.glGenFramebuffers(1, &framebuffer);
g_GLESFuncs.glBindFramebuffer(GL_FRAMEBUFFER, framebuffer);
g_GLESFuncs.glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, renderbuffer);
const GLenum status = g_GLESFuncs.glCheckFramebufferStatus(GL_FRAMEBUFFER);
if (status != GL_FRAMEBUFFER_COMPLETE) {
why = std::format("the host frame as a color attachment -> framebuffer status 0x{:04x}", status);
g_GLESFuncs.glDeleteFramebuffers(1, &framebuffer);
g_GLESFuncs.glDeleteRenderbuffers(1, &renderbuffer);
points.DestroyImage(g_EGLFuncs.eglGetCurrentDisplay(), image);
return false;
}
target.Image = image;
target.Renderbuffer = renderbuffer;
target.Framebuffer = framebuffer;
target.Width = desc.width;
target.Height = desc.height;
MGLOG_I("host frame target: %ux%d as framebuffer %u (stride %u, format 0x%x)", desc.width, desc.height,
framebuffer, desc.stride, desc.format);
return true;
}
void HostFrameTargetBind(const HostFrameTarget& target) {
if (target.Framebuffer == 0) return;
// THROUGH THE ENGINE'S SHADOW, never raw: a raw bind would leave the shadow claiming
// the previous framebuffer and false-skip the next re-bind (FramebufferImpl, Managers.h).
FramebufferImpl::BindFramebufferId(GL_FRAMEBUFFER, target.Framebuffer);
}
void HostFrameTargetDestroy(HostFrameTarget& target) {
if (target.Framebuffer != 0 && g_GLESFuncs.glDeleteFramebuffers)
g_GLESFuncs.glDeleteFramebuffers(1, &target.Framebuffer);
if (target.Renderbuffer != 0 && g_GLESFuncs.glDeleteRenderbuffers)
g_GLESFuncs.glDeleteRenderbuffers(1, &target.Renderbuffer);
if (target.Image != nullptr) {
String why;
const EntryPoints& points = Resolved(why);
if (points.DestroyImage != nullptr && g_EGLFuncs.eglGetCurrentDisplay)
points.DestroyImage(g_EGLFuncs.eglGetCurrentDisplay(), static_cast<EGLImageKHR>(target.Image));
}
target = HostFrameTarget{};
}
}
#else // !(MOBILEGL_BUILD_DISAGGREGATED && __ANDROID__)
namespace MobileGL::MG_Backend::DirectGLES {
Bool HostFrameTargetSupported(String& why) {
why = "host frames are a disaggregated Android path";
return false;
}
Bool HostFrameTargetCreate(const struct AHardwareBuffer*, HostFrameTarget&, String& why) {
why = "host frames are a disaggregated Android path";
return false;
}
void HostFrameTargetBind(const HostFrameTarget&) {}
void HostFrameTargetDestroy(HostFrameTarget& target) { target = HostFrameTarget{}; }
}
#endif
@@ -0,0 +1,52 @@
// MobileGL - MobileGL/MG_Backend/DirectGLES/HostFrameTarget.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Includes.h>
#include <EGL/egl.h>
#include <EGL/eglext.h>
struct AHardwareBuffer;
namespace MobileGL::MG_Backend::DirectGLES {
// THE FRAME THE DISPLAY HOST IS ABOUT TO SCAN OUT, AS A DRAW TARGET.
//
// The display host (the Android side that owns the screen) allocates its frames itself and
// offers each one over a socket as an AHardwareBuffer. That is not a convenience: measured
// on the device, AHardwareBuffer_createFromHandle accepts an AHardwareBuffer this side
// allocated and refuses one the display queue handed over (gralloc BAD_VALUE), so the frames
// a render server can draw into are the ones the host made for it.
//
// What this makes of such a frame: an EGLImage the driver accepts as a renderbuffer, and a
// framebuffer that is the drawing target while that frame is the current one. None of
// Android's window machinery is involved - no ANativeWindow, no SurfaceFlinger - which is
// the point: the host puts the frame on the glass when this side says it is drawn.
struct HostFrameTarget {
void* Image = nullptr; // EGLImageKHR; opaque here so this header stays free of EGL types
Uint Renderbuffer = 0;
Uint Framebuffer = 0;
Uint Width = 0;
Uint Height = 0;
};
// Whether the driver advertises what this is spelled in. False is answered with the missing
// name in `why`, never an abort: a server that cannot take host frames has to be able to
// say so.
Bool HostFrameTargetSupported(String& why);
// Imports `buffer` as a render target. On failure `target` is left empty and `why` says
// which call refused; no GL name is leaked.
Bool HostFrameTargetCreate(const struct AHardwareBuffer* buffer, HostFrameTarget& target, String& why);
// Binds the target as the current draw framebuffer, called when its frame becomes current.
void HostFrameTargetBind(const HostFrameTarget& target);
// Drops the GL names and the EGLImage. The AHardwareBuffer belongs to the bridge, which
// releases it; this never touches it.
void HostFrameTargetDestroy(HostFrameTarget& target);
}
File diff suppressed because it is too large Load Diff
+232 -10
View File
@@ -479,9 +479,29 @@ namespace MobileGL::MG_Backend::DirectGLES {
BackendPtr* GetOrCreateByHandle(MG_Pipe::MGPipeHandle handle) {
if (!EsprytSlotTablesEnabled()) return nullptr;
if (MG_Pipe::MGPipeHandleIsNull(handle)) return nullptr;
if (handle.Slot >= SlotTable::kMaxHandleSlot) return nullptr;
if (handle.Slot >= SlotTable::kMaxHandleSlot) {
#if MOBILEGL_BUILD_DISAGGREGATED
MG_Pipe::MGPipeSessionFail( // @Ph-declined (ID-P7-1): PH-2 stays Fatal, CONTRACT-P7 §12
MG_Pipe::MGPipeFatalFamily::ProtocolCorruption,
"MGPipe: Fatal{ProtocolCorruption, \"BackendSlotTable.HandleSlot\"} - "
"GetOrCreateByHandle named slot %u, past this table's %u bound",
handle.Slot, SlotTable::kMaxHandleSlot);
#else
return nullptr;
#endif
}
const Uint32 liveGen = m_slotTable.LiveGenAt(handle.Slot);
if (liveGen != 0 && liveGen > handle.Gen) return nullptr;
if (liveGen != 0 && liveGen > handle.Gen) {
#if MOBILEGL_BUILD_DISAGGREGATED
MG_Pipe::MGPipeSessionFail( // @Ph-declined (ID-P7-1): PH-2 stays Fatal, CONTRACT-P7 §12
MG_Pipe::MGPipeFatalFamily::ProtocolCorruption,
"MGPipe: Fatal{ProtocolCorruption, \"BackendSlotTable.Generation\"} - "
"GetOrCreateByHandle named generation %u at slot %u, behind live generation %u",
handle.Gen, handle.Slot, liveGen);
#else
return nullptr;
#endif
}
return &m_slotTable.GetOrCreate(handle);
}
@@ -687,12 +707,24 @@ namespace MobileGL::MG_Backend::DirectGLES {
// BackendSamplerObject* rows;
// programs - g_programTwinLookupMemo.
//
// THREE OF THEM CARRY A DEPENDENCY (MGPipe.h, D-K2) and it is diagnosed and REFUSED here
// rather than half-run, the bit-8-requires-bit-7 shape ResolveVertexInputSubsystemArm
// already ships: bit 11 requires bit 10, bit 9 requires bit 10, bit 10 requires bit 7. The
// mirror pairs (10 without 11, 10 without 9, 7 without 10) are all FINE and are said so out
// loud, because an unreachable branch that says something different is how the reachable
// one drifts. Bit 12 depends on nothing: a ShaderCso handle names no texture and no buffer.
// THE DEPENDENCY ROWS ARE IN MG_Pipe/SubsystemDeps.def, ONCE (P3b/P4b R-5), and the four
// resolvers below READ them through MGPipeSubsystemDependenciesAreSet rather than carrying a
// copy: a dependency the table declares is diagnosed and REFUSED here rather than half-run,
// the bit-8-requires-bit-7 shape ResolveVertexInputSubsystemArm already ships.
//
// THIS COMMENT USED TO RESTATE THE ROWS AND WAS WRONG IN TWO WAYS THAT NOTHING COULD FAIL,
// in the header of the file that IMPLEMENTS the refusal. It said "THREE OF THEM CARRY A
// DEPENDENCY" when there are six rows (bits 8 and 13 were the two it did not know about),
// and it listed "10 without 11" among the mirror pairs that are FINE - when 10-without-11 is
// precisely D-K2's FOURTH row (ID-14/ID-15), refused twenty lines below by
// ResolveTextureResourceSubsystemArm and withheld by the client at PipeFill.cpp's own gate.
// MGPipe.h carried the identical pair of defects and lost them with the table; this copy was
// left behind because Managers.h belonged to a package in flight. The rows are not restated
// here again: the file that states them is the one the code reads.
//
// The mirror pairs that really ARE fine are stated in MGPipe.h beside the table, because an
// absence is not a row - and because that is the one place where saying it cannot drift away
// from the rows it is the complement of.
//
// ResolveFramebufferSubsystemArm additionally carries D-C3's bring-up refusal: the wire
// array is MGPFramebufferState::Color[8] and GetDynamicParameters().MaxColorAttachments is
@@ -770,6 +802,48 @@ namespace MobileGL::MG_Backend::DirectGLES {
// composite: this reader hides the split behind one lookup, exactly as the wire does.
const MG_Pipe::MGPipeShaderCsoRecord* PipeShaderCsoRecordForHandle(MG_Pipe::MGPipeHandle cso);
// P3b/P4b (wave 2-D, package D3): THE STORAGE RECORD BEHIND A TEXTURE RECORD - the record of
// the texture it views for a glTextureView, and itself for every other texture.
//
// WHY THERE IS A SECOND READER. A view owns no texels, and the client keeps ONE emission
// cursor per storage: an upload through a view's own name is remapped onto its storage owner
// before it is emitted (MG_Impl/Pipe/TextureEmit.h, `ViewOf names the storage owner, and ONE
// HOP always reaches storage`), so every resource_subdata for either name is keyed on the
// OWNER and the applier moves the OWNER's Serial and PendingUploads alone
// (MG_Pipe/PipeApply.cpp, ApplyTextureUpload). A view therefore has a Serial that no upload
// ever moves, and the two STORAGE clauses of its draw-path clean gate have to be asked of
// this record instead of its own - which is what the monolith gate gets for free, because
// GetContentVersion() through a view is forwarded to the owner
// (MG_State/GLState/TextureState/TextureObjectView.cpp:100-102). CONTRACT-P5E.md §5.2 states
// the rule; without it a view that has been sampled once never sees another owner write.
//
// NULL IS "ASK AGAIN", NEVER "CLEAN". Every caller treats a null answer as a reason to RUN
// the sync, which is the direction that re-uploads rather than the direction that shows stale
// texels, and it raises nothing: a missing record is PipeTextureRecordForHandle's own
// null-on-miss answer one level up, not a protocol refusal.
//
// BOUNDED, THOUGH ONE HOP IS ALWAYS ENOUGH. glTextureView composes a view-of-a-view onto the
// ROOT at creation - the spec's additive min-level rule - and TextureObjectView's invariant
// is that its owner "is never a view itself", so Desc.ViewOf already names storage. The walk
// is written transitively anyway and gives up after a fixed number of hops, because THIS side
// may not depend on a client invariant to terminate: a chain the wire could make cyclic ends
// as a null answer rather than as a hang.
//
// DELETION ORDERING: THE LIVE/GEN TEST IS THE GUARANTEE, NOT THE EMIT ORDER. A view holds a
// strong SharedPtr to its storage owner (TextureObjectView::m_storageOwner) and the client
// emits resource_destroy from the DESTRUCTOR - "the last SharedPtr to this object dropping,
// not the glDelete* that only marks the name and leaves a still-bound object very much alive"
// (MG_State/GLState/TextureState/TextureObject.cpp:63). So the owner's storage cannot go away
// under a live view, whatever order the application deletes the two names in. But the two
// destroys are emitted back to back from one destructor chain, and when the view holds the
// owner's LAST reference the member is destroyed before the view's base destructor runs, so
// the wire carries resource_destroy(owner) and then resource_destroy(view): for one apply the
// view record is live with Desc.ViewOf naming a freed slot. No draw can land in between, and
// PipeTextureRecordForHandle tests Live and Gen, so that window answers null (= not clean)
// rather than a recycled stranger's record.
const MG_Pipe::MGPipeResourceRecord* PipeTextureStorageRecordForRecord(
const MG_Pipe::MGPipeResourceRecord& record);
// P5e (fb): the texture's own TARGET, from its descriptor. The image-unit bind needs it
// (glBindImageTexture's layered/format rules are per target) and MGPImageView has no room
// for it - 24 bytes, no pad - so it is read off the resource record instead of being added
@@ -930,6 +1004,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
// (the UBO binding walk) skip the probe while their pre-pass epoch read
// matches, exactly like the per-VAO memo stamps.
Uint64 drawCleanEpoch = 0;
// Frame serial whose completion retires the last glCopyBufferSubData the upload
// ring staged INTO this store (the flush ladder's ring tier: FlushPendingRangesNow
// on the pull arm, FlushPendingRangesFrom on the push arm); 0 = none. Until it
// retires, both keep every partial range off the range-invalidating map: on
// Adreno 830 that map, on a store with a queued copy, loses the copy (see
// FlushPendingRangesNow and InvalidateFlushAccessFor). Draw thread only, like the
// flush.
Uint64 ringCopyRetireSerial = 0;
// Zero-copy coherent persistent map (EXT_buffer_storage): the GL store is
// immutable, persistently+coherently mapped, and persistentPtr is what the app
// (and the frontend PipeResource) write into directly. While set, draw-time
@@ -968,6 +1050,21 @@ namespace MobileGL::MG_Backend::DirectGLES {
// the legacy arm did; this member exists for the drains that have no object, which
// in P3a is the readback flush and the fp64 narrowing.
const Uint8* hostBytes = nullptr;
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// P11 B2 (T0): the client's AHardwareBuffer this store IS - imported with
// glBufferStorageExternalEXT, so `id` is immutable, persistentMapped and
// persistentPtr is this server's own coherent map of the client's pages. The
// reference is this twin's own, released only after a fence that follows the
// store's last use has signalled (RetireT0Import). Survives an ES context loss:
// the next ensure re-imports it. Null for every store T0 did not adopt.
void* externalAhb = nullptr;
// P8-C: A SHADER WROTE THIS STORE ON THIS SIDE since the R-11 staged copy (or, for
// T0, the coherent map) was last known to match it - set by the draw / dispatch that
// bound it writable (storage block, atomic counter, writable buffer image), cleared
// by SplitHostBytesForCpuRead's refresh and by a respecify. Only the server's own
// CPU readers consult it; the client keeps its own set (GpuWritePending.h).
Bool serverGpuWritten = false;
#endif
};
@@ -1029,6 +1126,44 @@ namespace MobileGL::MG_Backend::DirectGLES {
// resolver. A no-op for a base that is not this resource's server shadow.
void RequireStagedCoverage(GLESBufferResource& resource, const Uint8* hostBase, SizeT start,
SizeT end, const char* site);
// P11 B2: THE CPU-READABLE BYTES OF A SPLIT SERVER'S BUFFER TWIN, for the handle-arm
// readers that have no frontend object (indirect commands, primitive-restart
// substitution, the multi-draw index rebase): the R-11 staged copy, or - for a T0 store,
// which has none - this server's own coherent map of the client's AHardwareBuffer, read
// exactly as the monolith arm reads the adopted map through MappedData().
inline const Uint8* SplitHostBytes(const GLESBufferResource& resource) {
if (resource.hostBytes != nullptr) return resource.hostBytes;
return resource.externalAhb != nullptr ? static_cast<const Uint8*>(resource.persistentPtr) : nullptr;
}
// P8-C: THE SAME BYTES, MADE CURRENT FIRST - for a reader that CONSUMES them (the restart
// rewrite, the multi-draw rebase, the *IndirectCount count, an indirect command's
// CPU-fed fields, the XFB scatter's pre-capture bytes). SplitHostBytes above is only
// "what uploads staged": a store a shader wrote on this side (serverGpuWritten) is
// refreshed from this server's own GL buffer - queued uploads flushed into it first,
// then the whole store read back into the staged copy, which also gives an orphaned
// store its first copy - and a T0 store lands its queued resident writes and finishes
// before its coherent map is read. A clean staged store costs one flag test.
// `res` names the record whose descriptor width the copy takes; `site` names the
// reader in the one-time diagnostic when the refresh cannot run.
const Uint8* SplitHostBytesForCpuRead(GLESBufferResource& resource, MG_Pipe::MGPipeHandle res,
const char* site);
// P8-C: bytes the SERVER itself put into a store's GL buffer and already holds on the CPU
// (an XFB capture it mapped for the client's writeback, a scatter it composed) go into
// the staged copy as well, so no refresh is owed for them. A T0 store has no copy.
void StageServerWrittenRange(GLESBufferResource& resource, MG_Pipe::MGPipeHandle res, const void* bytes,
SizeT offset, SizeT size);
// M-3's WHOLE-STORE predicate, and the GUARD every caller of RequireStagedCoverage over
// a whole-store read has to carry: true when the application DECLARED the store's
// content (glBufferData(size, data)), so a coverage gap is a missing record; false when
// it ORPHANED the store (glBufferData(size, NULL)), where every byte it has not staged
// since is undefined by its own declaration and the shadow's zero-fill is exactly what
// the monolith arm would have uploaded from MappedData(). Managers.cpp's pool-reuse
// ladder and DirectGLES.cpp's XFB scatter are its two callers and they must not be able
// to answer it differently.
Bool ResourceContentIsDeclared(MG_Pipe::MGPipeHandle res);
#endif
// MONOLITH GLUE, and named as such: the handle of a resource this backend is looking
@@ -1059,10 +1194,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
// ("do not take tier 1"), which drops the range onto the staging ring and costs a copy
// rather than a corruption.
//
// `ringCopyInFlight`: the store still has an upload-ring glCopyBufferSubData queued that the
// GPU may not have run (GLESBufferResource::ringCopyRetireSerial not yet retired).
//
// Returns the glMapBufferRange access bits, or 0 when tier 1 must not be taken.
inline constexpr SizeT kEsprytInvalidateRangeMinBytes = 128u * 1024u;
constexpr GLbitfield InvalidateFlushAccessFor(SizeT queuedStart, SizeT queuedEnd, SizeT mapStart,
SizeT mapEnd, SizeT limit, SizeT storageSize) {
SizeT mapEnd, SizeT limit, SizeT storageSize,
Bool ringCopyInFlight) {
if (mapEnd <= mapStart) return 0u;
// THE WIDENING REFUSAL. Not >=, not "covers": exactly, in both directions. A map
// narrower than the queued range leaves bytes unwritten inside a range it has just
@@ -1074,6 +1213,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
// page-substitution fast path needs a page-coverable range to engage, and below it
// the driver falls back to waiting out the WAR hazard on the CPU.
if (!wholeBuffer && size < kEsprytInvalidateRangeMinBytes) return 0u;
// A partial range-invalidating map must not overtake a ring copy into this store the GPU
// has not run yet. Adreno 830 answers such a map on a store with a copy still queued by
// losing the copy (dev f973008c, the pull arm's FlushPendingRangesNow: create-instancing
// drew whole Embeddium terrain sections empty, SSIM 0.870). The range goes to the ring
// instead, GPU-ordered behind the copies. The whole-buffer orphan-map is exempt: it
// rewrites every byte from the shadow the queued copies were taken from.
if (!wholeBuffer && ringCopyInFlight) return 0u;
return GL_MAP_WRITE_BIT |
(wholeBuffer ? GL_MAP_INVALIDATE_BUFFER_BIT : GL_MAP_INVALIDATE_RANGE_BIT);
}
@@ -1183,6 +1329,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
// Deletes GL buffers whose owning frontend objects died (possibly on a
// thread without a current ES context). Called from draw-time sync.
void ProcessDeferredBufferReleases();
#if MOBILEGL_BUILD_DISAGGREGATED
// P11 B2: the T0 half of that drain - run right after it (retired AHardwareBuffers).
void ProcessDeferredT0Retires();
#endif
// glBindBuffer with a redundant-bind cache for GL_ARRAY_BUFFER.
void BindBufferId(GLenum target, Uint id);
@@ -1729,6 +1879,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
return TextureTarget::Texture2D;
case TextureTarget::Texture1DArray:
return TextureTarget::Texture2DArray;
// An external texture's storage is an imported EGLImage; to the driver it is 2D.
case TextureTarget::TextureExternalOES:
return TextureTarget::Texture2D;
default:
return target;
}
@@ -1879,7 +2032,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
// allocation, or capture the old native contents, merge pending
// upload regions and re-arm the server's whole-level replay set.
// First allocations use staged bytes directly. No frontend dirty
// flag, allocator or pixel-store object participates.
// flag, allocator or pixel-store object participates. P9 W2: a level
// the driver does not hold (defined after the last sync) is never read
// back, and a readback the driver refuses falls back to the staged
// level when the store covers it (notes/p9/W2-REMINT.md).
void RequireImageBindableStorageByHandle(MG_Pipe::MGPipeHandle res,
const MG_Pipe::MGPipeResourceRecord& record);
#endif
@@ -1888,6 +2044,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
// carrier the storage is an integer texture holding codes and glGetTexImage still owes
// the application floats.
Bool RequiresImageBindableStorage() const { return m_imageBindableStorageRequired; }
// See m_driverMayHoldGpuWrites. Callers holding the STATE texture use
// NoteDriverSideTextureWrite, which also reaches the storage behind a view.
void NoteDriverSideWrite() { m_driverMayHoldGpuWrites = true; }
void Bind(GLenum target, Uint unit = TempTextureUnit);
Uint GetBackendTextureId() const;
@@ -2004,6 +2163,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
Bool m_isInitialized = false;
Bool m_imageBindableStorageRequired = false;
Bool m_backendStorageImmutable = false;
// Whether the driver texture may hold texels the shadow never saw. Set the first time
// anything other than a texel upload can write it - an application framebuffer
// attachment, a writable image binding, a copy or blit into it, a driver-side mipmap
// generation - and never cleared. Until then the driver holds, outside a level's
// pending writes, exactly the shadow bytes it was last sent, so those writes may go up
// as their union box (one unpack-ring job, which is what Mali needs); from then on
// only the texels the client wrote may go up, since the shadow is stale everywhere
// else. See SyncMipmapsToBackend.
Bool m_driverMayHoldGpuWrites = false;
// Latches the "this driver has no buffer textures" report to once per texture. The
// report is emitted from the respecify path, which bails before recording the state
// it was asked to apply - so without the latch the texture stays permanently dirty
@@ -2173,6 +2341,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
SharedPtr<BackendTextureObject>& SyncTextureObjectToBackend(
const SharedPtr<MG_State::GLState::ITextureObject>& textureObject,
Bool imageBindableStorageRequired = false);
// Every path that lets the DRIVER write a texture's texels calls this (see
// BackendTextureObject::m_driverMayHoldGpuWrites): an application framebuffer attachment,
// a writable image binding, a copy, blit or mipmap generation into it. A view writes its
// storage texture's image, so the storage texture is the one marked.
void NoteDriverSideTextureWrite(const SharedPtr<MG_State::GLState::ITextureObject>& textureObject);
#if MOBILEGL_PIPE_PUSH
// ---- P5e SEAM (MG_Remote/CONTRACT-P5E.md §4.2, §5.2; declared by c0e, bodied by
// id/tx2) --------------------------------------------------------------------------
@@ -2189,6 +2362,29 @@ namespace MobileGL::MG_Backend::DirectGLES {
SharedPtr<BackendTextureObject>& SyncTextureToBackendByHandle(
MG_Pipe::MGPipeHandle texture, Bool imageBindableStorageRequired = false);
BackendTextureObject* ResolveTextureTwin(MG_Pipe::MGPipeHandle texture);
// NoteDriverSideTextureWrite's HANDLE ARM, for the same paths when the caller holds a
// Texture handle (a framebuffer surface's Res, an image unit's Res, VerbMipRes,
// VerbCopyTexDst, a copy-image endpoint): the twin the handle's record names is marked, or
// its storage owner's (Desc.ViewOf) for a view. Nothing on the apply thread reads a
// frontend object for it.
//
// P8-E: under a transport the same storage owner's staged store is marked driver-written
// too (StagedTextureStore::MarkDriverWritten), so a refused readback does not answer from
// bytes the driver has since overwritten - unless `storeFollowsTheWrite`, which only
// glCopyImageSubData passes: it tries FollowCopyImageInStagedStore first and marks the store
// itself when that fails.
void NoteDriverSideTextureWriteByHandle(MG_Pipe::MGPipeHandle texture, Bool storeFollowsTheWrite = false);
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// P8-E (notes/p8/E.md): make glCopyImageSubData's texel-block move in the server's staged
// store too (StagedTextureStore::FollowCopy), so the destination level's bytes stay its
// content. Box in each image's own texel grid. False when the store cannot follow: a cube
// map or 1D array endpoint (copy-image puts faces / layers on z, the store does not), a view,
// two different internal formats, a level whose bytes did not all cross, a source the
// driver has written, or a monolith arm.
Bool FollowCopyImageInStagedStore(MG_Pipe::MGPipeHandle source, Uint32 sourceLevel, const IntVec3& sourceOrigin,
MG_Pipe::MGPipeHandle destination, Uint32 destinationLevel,
const IntVec3& destinationOrigin, const IntVec3& size);
#endif
// Brings every texture the next draw reads - the touched units' bindings and the draw
// FBO's texture attachments - onto the backend, through the two borrowed-pair memos
@@ -2946,6 +3142,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
// Same for gl_BaseVertex: only a program that reads it pays for the per-draw
// uniform write, and only such a program needs the reset after one.
Bool ReadsBaseVertex() const { return m_baseVertexUniformLocation >= 0; }
#if MOBILEGL_BUILD_DISAGGREGATED
// P8-C: same for the gl_BaseInstance uniform - which an indirect draw feeds from the
// command on the CPU only when the mg_IndirectParams view is absent.
Bool ReadsBaseInstance() const { return m_baseInstanceUniformLocation >= 0; }
#endif
// Which viewport indices the next draw's fragments may keep, one bit each. Written
// once per replay pass; see ForEachViewportRoutingPass.
void SetViewportPassMask(Uint32 indexMask) const;
@@ -3524,5 +3725,26 @@ namespace MobileGL::MG_Backend::DirectGLES {
// delete opcode may already have released (the idempotent-second-path shape the notice
// arms document).
Bool ReleaseTwinsForWireObjectDeath(MG_Pipe::MGPipeHandle handle, MG_Pipe::MGPipeKind kind);
// P12 (on-screen server window): A SERVER SESSION ENDS IN A PROCESS THAT OUTLIVES IT.
//
// The twin tables above are process globals, and until P12 every server process served ONE
// session: the exec'd supervisor forks a child per session and the child exits with it, so the
// twins a session built died with its process (InProcessTeardown's deliberate leak). The
// in-process display server (mobilegl_server_serve_inprocess, the display Activity's process)
// runs sessions ONE AFTER ANOTHER IN ONE PROCESS, and the next session's client mints its
// handles from the same {slot, gen} space again. A twin the previous session left behind then
// answers for the new session's object: a program twin names a program of the DESTROYED
// context, so the link fails with an empty log and every draw no-ops (Espryt on Adreno and on
// llvmpipe alike), a stale VAO/buffer twin feeds the driver a dead name (a SIGSEGV inside
// Adreno's glDrawElements), and a stale live generation ahead of the new client's refuses its
// handle as ProtocolCorruption.
//
// So the server backend's destruction - the end of a session under a transport, on the apply
// thread, after DestroyEGLContext destroyed the context those ids belonged to - drops EVERY
// twin of every kind, with InProcessTeardown() answering true for the duration so no twin
// destructor calls into the driver (there is no current context, and the ids name nothing).
// A no-op once exit() has begun (the statics may already be gone). Monolith never calls it.
void DropEveryTwinForEndedServerSession();
#endif
} // namespace MobileGL::MG_Backend::DirectGLES
+7 -6
View File
@@ -191,15 +191,16 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
if (question == IndexBufferQuestion::Presence) return view;
if (question == IndexBufferQuestion::HostBytes) {
// No Sync* pair here and none is missing: on this side the applier has
// already consumed the persistent-map blocks and the shader writebacks for
// this resource before the draw verb replayed, so the server's staged shadow
// IS the synced copy. That is also why this arm cannot be expressed as "find
// the object and run the monolith body".
// The applier has already consumed the persistent-map blocks for this
// resource before the draw verb replayed; the one Sync* this arm owes is the
// monolith's SyncGpuWrites, and (P8-C) SplitHostBytesForCpuRead is it: a store
// a shader wrote on this side is read back from this server's own GL buffer,
// because no writeback ever reaches the server's staged copy. That is also why
// this arm cannot be expressed as "find the object and run the monolith body".
auto* resource = BufferImpl::FindBufferResourceForHandle(res);
if (resource == nullptr) RefuseMissingIndexBufferRecord(entry, res, "backend resource");
view.Size = BufferImpl::ResourceWidthForHandle(res);
view.HostBytes = resource->hostBytes;
view.HostBytes = BufferImpl::SplitHostBytesForCpuRead(*resource, res, "multidraw_index_rebase");
if (view.HostBytes != nullptr && IndexBufferRecordHasDefinedContent(st, res)) {
BufferImpl::RequireStagedCoverage(*resource, view.HostBytes, 0, view.Size,
"multidraw_index_rebase");
@@ -14,6 +14,9 @@
#if MOBILEGL_PIPE_PUSH
#include <MG_Impl/Pipe/SlotAllocator.h>
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Pipe/PipeSessionFail.h>
#endif
// Espryt 0b, the first Track H slice: the DENSE, {slot, gen}-keyed twin table that replaces
// StateBackendObjectRegistry's UnorderedMap<StateObject*, Entry>.
@@ -345,9 +348,19 @@ namespace MobileGL::MG_Backend::DirectGLES {
// already orders of magnitude past any real GL object count, while a corrupt 32-bit
// slot asks for a four-billion-entry resize.
if (handle.Slot >= kMaxHandleSlot) {
#if MOBILEGL_BUILD_DISAGGREGATED
// PH-2: handle.Slot is peer supplied on the disaggregated arm. Release builds
// must publish a named protocol fault instead of silently returning m_nullTwin.
MG_Pipe::MGPipeSessionFail( // @Ph-declined (ID-P7-1): PH-2 stays Fatal, CONTRACT-P7 §12
MG_Pipe::MGPipeFatalFamily::ProtocolCorruption,
"MGPipe: Fatal{ProtocolCorruption, \"BackendSlotTable.HandleSlot\"} - "
"GetOrCreate(handle) named slot %u, past this table's %u bound",
handle.Slot, kMaxHandleSlot);
#else
MOBILEGL_ASSERT(false, "GetOrCreate(handle) named slot %u, past this table's %u bound",
handle.Slot, kMaxHandleSlot);
return m_nullTwin;
#endif
}
// Same arming as the minting overload, and for the same reason: twin creation is
@@ -369,11 +382,21 @@ namespace MobileGL::MG_Backend::DirectGLES {
// below already gives the same input.
Entry& entry = EntryAt(handle.Slot);
if (entry.Live && entry.Gen > handle.Gen) {
#if MOBILEGL_BUILD_DISAGGREGATED
// PH-2: a stale peer generation must not silently shed the incumbent twin in a
// release server. The funnel names the exact identity fault and notifies the peer.
MG_Pipe::MGPipeSessionFail( // @Ph-declined (ID-P7-1): PH-2 stays Fatal, CONTRACT-P7 §12
MG_Pipe::MGPipeFatalFamily::ProtocolCorruption,
"MGPipe: Fatal{ProtocolCorruption, \"BackendSlotTable.Generation\"} - "
"GetOrCreate(handle) named generation %u at slot %u, behind live generation %u",
handle.Gen, handle.Slot, entry.Gen);
#else
MOBILEGL_ASSERT(false,
"GetOrCreate(handle) named generation %u at slot %u, which is BEHIND "
"the live entry's %u - refusing rather than destroying the incumbent",
handle.Gen, handle.Slot, entry.Gen);
return m_nullTwin;
#endif
}
if (entry.Live && entry.Gen != handle.Gen) entry.backend.reset();
entry.Gen = handle.Gen;
@@ -1,6 +1,7 @@
// Included in DirectGLES.cpp after the native packed-word readback helper.
// The server needs owned tight rows for a storage promotion, not client PACK
// layout or a fallback into a frontend texture's potentially stale shadow.
// A storage promotion needs the driver's own texels as owned tight rows, not
// client PACK layout or a fallback into a frontend texture's potentially stale
// shadow.
class ScopedWireTextureReadLevel {
public:
ScopedWireTextureReadLevel(GLuint texture, GLenum target, GLint level)
@@ -280,6 +281,144 @@
return true;
}
#if MOBILEGL_BUILD_DISAGGREGATED
// P9 W2-a. The store's Defined-ness follows the client's respecifies; the driver's follows the
// twin's last sync. A re-mint asks this before reading a level back, because a level defined
// after that sync (null data, or null data plus a partial upload) is not on the driver yet and
// its FBO attachment is incomplete - which used to end the session at
// Fatal{ResourceUnavailable, "image-promotion-readback"} on legal GL.
Bool NativeTextureLevelHasExtent(GLuint texture, TextureTarget target, TextureUploadTarget uploadTarget,
GLint level, const IntVec3& logicalExtent) {
if (!texture || level < 0 || !g_GLESFuncs.glGetTexLevelParameteriv) return true;
const IntVec3 extent = TextureImpl::GetBackendUploadSize(target, logicalExtent);
const GLenum nativeTarget = TextureImpl::ConvertTextureTargetToBackendGLEnum(target);
GLenum binding = GL_NONE;
switch (nativeTarget) {
case GL_TEXTURE_2D: binding = GL_TEXTURE_BINDING_2D; break;
case GL_TEXTURE_3D: binding = GL_TEXTURE_BINDING_3D; break;
case GL_TEXTURE_2D_ARRAY: binding = GL_TEXTURE_BINDING_2D_ARRAY; break;
case GL_TEXTURE_CUBE_MAP: binding = GL_TEXTURE_BINDING_CUBE_MAP; break;
case GL_TEXTURE_CUBE_MAP_ARRAY: binding = GL_TEXTURE_BINDING_CUBE_MAP_ARRAY; break;
default: return true;
}
// A cube's levels are per face; every other target answers for the whole level.
const GLenum queryTarget = nativeTarget == GL_TEXTURE_CUBE_MAP
? TextureImpl::ConvertTextureUploadTargetToBackendGLEnum(uploadTarget) : nativeTarget;
DrainESErrors();
GLint previous = 0;
g_GLESFuncs.glGetIntegerv(binding, &previous);
g_GLESFuncs.glBindTexture(nativeTarget, texture);
GLint width = 0, height = 0, depth = 0;
g_GLESFuncs.glGetTexLevelParameteriv(queryTarget, level, GL_TEXTURE_WIDTH, &width);
g_GLESFuncs.glGetTexLevelParameteriv(queryTarget, level, GL_TEXTURE_HEIGHT, &height);
g_GLESFuncs.glGetTexLevelParameteriv(queryTarget, level, GL_TEXTURE_DEPTH, &depth);
const Bool answered = g_GLESFuncs.glGetError() == GL_NO_ERROR;
g_GLESFuncs.glBindTexture(nativeTarget, static_cast<GLuint>(previous));
if (!answered) return true;
const Bool layered = nativeTarget == GL_TEXTURE_3D || nativeTarget == GL_TEXTURE_2D_ARRAY ||
nativeTarget == GL_TEXTURE_CUBE_MAP_ARRAY;
return width == extent.x() && height == extent.y() && (!layered || depth == extent.z());
}
// P8-E (notes/p8/E.md). MGITEST_ESPRYT_REFUSE_TEXTURE_READBACK_EXTENT=<w>x<h>, READ BY THE SERVER:
// a colour level of exactly that extent is treated as one the driver refused to attach - what an
// Adreno does for RGB16 and the 16-bit SNORM formats, and what no host driver can be made to do.
// Scoped by the extent so the tcp lane's shared supervisor can carry it for the whole lane: only
// the scenario that builds that extent (CopyImageStoreReadbackScenario) ever meets it.
static Bool WireTextureReadbackRefusedForTesting(const IntVec3& extent) {
static const IntVec3 s_refused = [] {
const char* value = std::getenv("MGITEST_ESPRYT_REFUSE_TEXTURE_READBACK_EXTENT");
if (value == nullptr) return IntVec3{0, 0, 0};
char* end = nullptr;
const long width = std::strtol(value, &end, 10);
if (end == value || *end != 'x') return IntVec3{0, 0, 0};
const long height = std::strtol(end + 1, nullptr, 10);
if (width <= 0 || height <= 0) return IntVec3{0, 0, 0};
return IntVec3{static_cast<Int>(width), static_cast<Int>(height), 1};
}();
return s_refused.x() > 0 && extent == s_refused;
}
// P8-E: THE STORE ROUTE, for a level the driver refused to read back. The server's staged store
// is the only CPU copy of the level on this side, and it answers only while its bytes ARE the
// level: every byte crossed (IsCovered), at the extent asked for, and the texture was never
// written by the driver except by copies the store followed (!IsDriverWritten - see
// StagedTextureStore's P8-E block and FollowCopyImageInStagedStore). The conversion is
// GetTexImageViaShadowConversion's, into tight rows: a packed format read with its own type
// takes the stored words verbatim, everything else decodes to wide RGBA and converts row by row.
// Anything else DECLINES BY NAME: the reply is an error, and the client's
// Fatal{ReplyError, "ReadPixels"} ends the session instead of handing back bytes the driver has
// since overwritten.
static Bool ReadWireLevelFromStagedStore(MG_Pipe::MGPipeHandle res, TextureInternalFormat logicalFormat,
TextureUploadTarget uploadTarget, GLint level, const IntVec3& extent,
GLenum format, GLenum type, Vector<Uint8>& bytes) {
auto& store = MG_Remote::Server::ServerStagedTexture();
const Uint64 key = MG_Remote::Server::StagedTextureStore::KeyForHandle(res);
const auto target = static_cast<Uint16>(uploadTarget);
const auto mip = static_cast<Uint16>(level);
ReadbackChannelMapping mapping{};
const SizeT pixelBytes = GetReadbackChannelMapping(format, mapping) ? GetReadbackDstPixelSize(mapping, type) : 0;
const SizeT texels = static_cast<SizeT>(std::max(extent.x(), 0)) * static_cast<SizeT>(std::max(extent.y(), 0)) *
static_cast<SizeT>(std::max(extent.z(), 0));
const char* declined = nullptr;
if (!store.CopiesIntoServerStorage()) {
declined = "this arm keeps no staged store";
} else if (store.IsDriverWritten(key)) {
declined = "the driver has written the texture since its bytes crossed";
} else if (!store.IsCovered(key, target, mip)) {
declined = "not every byte of the level crossed";
} else if (store.LevelExtentOrUndefined(key, target, mip) != extent) {
declined = "the staged level is not the extent asked for";
} else if (pixelBytes == 0 || texels == 0) {
declined = "the format/type pair has no readback mapping";
}
Vector<Uint8> out;
if (declined == nullptr) {
const Uint8* staged = store.RequireLevelBytes(key, target, mip, "texture-readback-from-store");
if (MG_Util::PixelStoreProcessor::IsRawPackedPixelTransfer(
logicalFormat, MG_Util::ConvertGLEnumToTextureInputFormat(format),
MG_Util::ConvertGLEnumToTexturePixelDataType(type))) {
if (store.LevelByteSize(key, target, mip) != texels * pixelBytes) {
declined = "the staged words are not the client's word size";
} else {
out.assign(staged, staged + texels * pixelBytes);
}
} else {
Vector<Uint8> wide;
Bool isInteger = false;
Bool isSigned = false;
if (!MG_Util::PixelStoreProcessor::DecodeShadowDataToWideRGBA(logicalFormat, staged, texels, wide,
isInteger, isSigned)) {
declined = "the staged format has no wide decoding";
} else if (mapping.isInteger != isInteger) {
declined = "an integer read of a normalized level, or the reverse";
} else {
const GLenum wideType = isInteger ? (isSigned ? GL_INT : GL_UNSIGNED_INT) : GL_FLOAT;
const SizeT rowTexels = static_cast<SizeT>(extent.x());
const SizeT rows = texels / rowTexels;
const SizeT wideRowBytes = rowTexels * 4 * GetReadbackComponentSize(wideType);
out.resize(texels * pixelBytes);
for (SizeT row = 0; row < rows; ++row) {
ReadbackImpl::ConvertWideReadbackRow(wide.data() + row * wideRowBytes,
out.data() + row * rowTexels * pixelBytes, rowTexels,
wideType, mapping, type);
}
}
}
}
if (declined != nullptr) {
MGLOG_E_ONCE("MGPipe: texture {%u,%u} level %d: the driver refused the readback and the staged store "
"cannot answer it (%s) - DECLINED (texture-readback-store-declined)",
res.Slot, res.Gen, level, declined);
return false;
}
MGLOG_W_ONCE("MGPipe: texture {%u,%u} level %d: the driver refused the readback (internal format %u); "
"answered from the server's staged store (texture-readback-from-store)",
res.Slot, res.Gen, level, static_cast<unsigned>(logicalFormat));
bytes = Move(out);
return true;
}
Bool ReadTextureImageWire(const MG_Pipe::MGPReadbackInfo& info, Vector<Uint8>& bytes) {
const auto* record = PipeTextureRecordForHandle(info.Res);
if (!record) return false;
@@ -293,17 +432,32 @@
const Bool wholeCube = info.Target == GL_TEXTURE_CUBE_MAP;
const IntVec3 extent{static_cast<Int>(info.Box.W), static_cast<Int>(info.Box.H),
wholeCube ? 1 : static_cast<Int>(info.Box.D)};
const auto logicalFormat = static_cast<TextureInternalFormat>(record->Desc.InternalFormat);
// The R32UI word leg never goes through a framebuffer, so the knob leaves it alone (it is
// ReadTextureLevelTight's first choice for exactly these reads).
const Bool packedWordRead =
!twin->RequiresImageBindableStorage() &&
MG_Util::PixelStoreProcessor::IsRawPackedPixelTransfer(
logicalFormat, MG_Util::ConvertGLEnumToTextureInputFormat(info.Format),
MG_Util::ConvertGLEnumToTexturePixelDataType(info.Type)) &&
MG_Util::GetInputBytesPerPixel(MG_Util::ConvertGLEnumToTextureInputFormat(info.Format),
MG_Util::ConvertGLEnumToTexturePixelDataType(info.Type)) == sizeof(Uint32);
const Bool refusedForTesting = !packedWordRead && WireTextureReadbackRefusedForTesting(extent);
Vector<Uint8> result;
for (Int face = 0; face < (wholeCube ? 6 : 1); ++face) {
const auto uploadTarget = wholeCube
? static_cast<TextureUploadTarget>(static_cast<Int>(TextureUploadTarget::CubeMapPositiveX) + face)
: MG_Util::ConvertGLEnumToTextureUploadTarget(info.Target);
Vector<Uint8> slice;
if (!ReadTextureLevelTight(twin->GetBackendTextureId(), target, uploadTarget,
static_cast<TextureInternalFormat>(record->Desc.InternalFormat), info.Level, extent,
twin->RequiresImageBindableStorage(), info.Format, info.Type, slice)) return false;
if (refusedForTesting ||
!ReadTextureLevelTight(twin->GetBackendTextureId(), target, uploadTarget, logicalFormat, info.Level,
extent, twin->RequiresImageBindableStorage(), info.Format, info.Type, slice)) {
if (!ReadWireLevelFromStagedStore(info.Res, logicalFormat, uploadTarget, info.Level, extent,
info.Format, info.Type, slice)) return false;
}
result.insert(result.end(), slice.begin(), slice.end());
}
bytes = Move(result);
return true;
}
#endif // MOBILEGL_BUILD_DISAGGREGATED
@@ -330,7 +330,29 @@ namespace MobileGL::MG_Backend::DirectVulkan {
PopulateFormatCapabilitiesImpl(physicalDevice, getFormatProperties, capabilities, cache);
}
#if MOBILEGL_BUILD_DISAGGREGATED
BackendObject_DirectVulkan::~BackendObject_DirectVulkan() {
// P12 review fix: A SERVER SESSION THAT ENDS WITHOUT A CLEAN TEARDOWN LEAVES NO RENDERER ON
// THE WINDOW. The renderer - swapchain and VkSurfaceKHR on the window - is a process global
// that only ReleaseEGLResources / a released surface drops, and a client that went away
// without eglTerminate (a crash, a kill, a dropped connection) sends neither. Under a
// transport this object is the server's and its destruction is the session's end (ServerLoop's
// apply thread, after the final drain, before the display lease ends), so the renderer goes
// here: nothing of the ended session keeps presenting into, or holding a surface on, the
// server's window once its lease is over. Monolith keeps its renderer.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith && pVulkanRenderer != nullptr) {
BumpRendererGeneration();
pVulkanRenderer.reset();
ClearProgramResourceCaches();
}
// And the swap interval it asked for (DirectVulkan.h).
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
ForgetRequestedSwapInterval();
}
}
#else
BackendObject_DirectVulkan::~BackendObject_DirectVulkan() = default;
#endif
BackendObject_DirectVulkan::BackendObject_DirectVulkan() : m_rendererInfo{GetRendererIdentity()} {}
@@ -345,7 +367,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Any renderer instance this assignment replaces is destroyed here;
// fence/timer-query handles stamped with the old generation go stale.
BumpRendererGeneration();
pVulkanRenderer = MakeUnique<MG_Backend::DirectVulkan::VulkanRenderer>(nativeWindow);
VulkanRendererConfig config;
config.SwapInterval = GetRequestedSwapInterval();
pVulkanRenderer = MakeUnique<MG_Backend::DirectVulkan::VulkanRenderer>(nativeWindow, config);
MOBILEGL_ASSERT(pVulkanRenderer != nullptr, "InitWindowSurface: VulkanRenderer creation failed");
pVulkanRenderer->Initialize();
return true;
@@ -367,6 +391,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void BackendObject_DirectVulkan::Initialize() {
#if MOBILEGL_BUILD_DISAGGREGATED
VkBufferManager::RegisterWireResourceOps();
#if MOBILEGL_PIPE_PUSH
// P7 wave 2 package C (CONTRACT-P7 §5.5). BESIDE the resource ops and for the same
// reason: this is the one place both Magma server roles pass through - the inproc
// role and the spawn/TCP session child - and ServerLoop::CreateBackend calls it at
// step 1 of InitServerRoleCommon, before the client session exists and therefore
// before any frontend object can die unheard. A per-transport install is how one of
// the two roles ends up without it.
InstallStateObjectDeathOps();
#endif
#endif
m_initialized = true;
}
@@ -459,6 +492,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return BackendObject::SwapEGLBuffers(dpy, draw);
}
void BackendObject_DirectVulkan::SetEGLSwapInterval(Int interval) {
// Same lock as SwapEGLBuffers: the renderer reads the request in Present.
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
BackendObject::SetEGLSwapInterval(interval);
}
void BackendObject_DirectVulkan::ReleaseEGLSurface(EGLSurface surface) {
const std::lock_guard<std::recursive_mutex> lock(m_eglStateMutex);
BackendObject::ReleaseEGLSurface(surface);
@@ -620,7 +659,38 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Advertised with GL_NUM_PROGRAM_BINARY_FORMATS = 0, which the
// extension explicitly permits. It is also the only thing that
// exposes glProgramParameteri before GL 4.1.
E_GL_ARB_get_program_binary};
E_GL_ARB_get_program_binary,
// kwin_wayland 6.7.3 will not use a context it cannot drive, and it asks for these
// three names first: EglContext::checkSupported() (src/opengl/eglcontext.cpp) takes the
// DESKTOP-OpenGL branch for every context whose GL_VERSION is not an "OpenGL ES"
// string, and that branch requires GL_ARB_shader_objects, GL_ARB_fragment_shader and
// GL_ARB_vertex_shader. GL_VERSION here is "4.6.0 MobileGL ...", so with the names
// missing checkSupported() answered false, EglContext::create() returned nullptr,
// EglBackend::createContext() returned false, and kwin printed "Could not initialize
// rendering context" and quit before drawing a frame.
//
// The three are true statements about this backend too, and the entry points behind
// them are the same ones the DirectGLES half of this list is advertised with: the
// frontend implements all 39 ARB_shader_objects procedures, ARB_vertex_shader's three
// attrib-location entry points, and nothing for ARB_fragment_shader, whose "New
// Procedures and Functions" section reads "None". The shader and program objects live in
// MG_Impl/MG_State, above the backend that draws with them, so a name that is true on
// one backend is true on the other.
E_GL_ARB_shader_objects, E_GL_ARB_vertex_shader, E_GL_ARB_fragment_shader,
// NPOT textures: Vulkan puts no power-of-two restriction on an image, its mip chain or
// its samplers, so this is the strongest form of the promise the string makes. Kept
// identical to the DirectGLES group so the two backends do not disagree about what
// MobileGL is.
E_GL_ARB_texture_non_power_of_two,
// GL_OES_EGL_image: the entry points are real (MG_Impl/GLImpl/Exporting/Definitions.cpp)
// and this name is what makes them reachable - libepoxy in kwin resolves
// glEGLImageTargetTexture2DOES BY NAME and refuses the import path without it, printing
// "No provider of glEGLImageTargetTexture2DOES found. Requires one of: GL_OES_EGL_image".
// That import path is how the compositor's scene reaches the display daemon's dma-bufs.
// The render-server half of the import (fd over the GBM socket, then the server-side
// EGLImage) is the next change; until it lands an imported texture has no storage behind
// it, which is logged once at the entry point rather than left to look like a black frame.
E_GL_OES_EGL_image};
// Vulkan's drawIndirectFirstInstance feature is optional. Direct base-instance calls work
// without it, but ARB_base_instance also promises non-zero firstInstance in GPU indirect
// commands; the renderer supplies true only when that word is legal and gl_InstanceID can
@@ -700,6 +770,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
static Bool funcsTableInitialized = false;
if (!funcsTableInitialized) {
funcsTable.Present = Present;
funcsTable.SetSwapInterval = SetSwapInterval;
funcsTable.GL.DrawArrays = DrawArrays;
funcsTable.GL.DrawElements = DrawElements;
funcsTable.GL.DrawElementsBaseVertex = DrawElementsBaseVertex;
@@ -34,6 +34,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool CreateEGLPbufferSurface(EGLSurface surface, EGLint width, EGLint height) override;
Bool MakeEGLCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext ctx) override;
Bool SwapEGLBuffers(EGLDisplay dpy, EGLSurface draw) override;
void SetEGLSwapInterval(Int interval) override;
void ReleaseEGLSurface(EGLSurface surface) override;
void ReleaseEGLResources() override;
@@ -0,0 +1,29 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/DescriptorPoolCensus.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <cstdint>
namespace MobileGL::MG_Backend::DirectVulkan {
// A count of Magma's per-frame descriptor pools and the sets they hold (UniformManager), for
// tests. Dependency-free on purpose: the integration scenarios include it next to the system GL
// headers. They link the static library on desktop; the shipping shared library is built with
// hidden visibility, so on Android nothing outside the library can call GetDescriptorPoolCensus.
struct DescriptorPoolCensus {
bool available = false; // false when no Magma renderer is live (e.g. the DirectGLES backend)
uint32_t frameSlots = 0;
uint32_t pools = 0; // descriptor pools across every frame slot
uint32_t maxPoolsInOneSlot = 0;
uint64_t capacitySets = 0; // sum of maxSets over those pools
uint64_t cachedSets = 0; // sets allocated from them and held in the per-layout caches
uint64_t slotTrims = 0; // whole-slot trims since the renderer came up
};
DescriptorPoolCensus GetDescriptorPoolCensus();
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -21,6 +21,13 @@
#include <Config.h>
#include <MG_Remote/Server/ServerLoop.h>
#include <MG_Pipe/PipeApply.h>
#if MOBILEGL_PIPE_PUSH
// P7 wave 2 package C (CONTRACT-P7 §5.5): Magma's own death-notice table. Both headers are
// push-only and reached here for the same reason DirectGLES/Managers.cpp reaches them - the
// notice is declared by the frontend and answered by whichever backend is running.
#include <MG_Remote/Client/WireTables.h>
#include <MG_State/GLState/StateObjectDeathNotice.h>
#endif
#endif
#include <atomic>
#include <bit>
@@ -55,6 +62,66 @@ namespace MobileGL::MG_Backend::DirectVulkan {
std::atomic<Uint64> g_rendererGeneration{1};
} // namespace
#if MOBILEGL_BUILD_DISAGGREGATED && MOBILEGL_PIPE_PUSH
namespace {
// P7 wave 2 package C, CONTRACT-P7 §5.5: MAGMA'S MIRROR OF g_glesStateObjectDeathOps
// (DirectGLES/Managers.cpp:335), and the one arm of that table Magma has any work in.
//
// WHAT THE NOTICE IS FOR HERE, and it is not what it is for on Espryt. Espryt's arm
// destroys a BACKEND TWIN keyed by lifetime id; Magma keeps no such twin - its two
// client-minted kinds (VertexElementsCso, Buffer) are keyed on {slot, gen} by
// MagmaPipeIdentityTables and every other kind is still reached from its frontend
// object. So the switch Espryt runs under `#else` has nothing to do on this backend
// and this table is exactly the EMIT arm: with a transport up and off the apply
// thread, the death crosses as the `object_death` record
// (MG_Remote::Client::EmitObjectDeathRecord, CONTRACT-P5C §5.2).
//
// WHY THAT MATTERS, measured rather than argued (package L's probe,
// notes/p7/magma-two-process-first-run.md §6): under SPAWN the client process has no
// DirectVulkan backend at all, so InstallClientWireTables' own emitter already
// answered the notice and CtWireScenario's two death cases PASSED. Under INPROC the
// server role's backend is a thread of this process, Transport is Split rather than
// Spawn, and WireTables' install condition deliberately stands aside for "the
// backend's own dispatcher" - which on Magma did not exist. Both cases therefore
// FAILED on inproc with `deaths` stuck at 0. This table is that dispatcher.
//
// FRAMEBUFFER IS WHY THE GAP WAS NOT MERELY COSMETIC. The client-side SLOT is freed
// backend-neutrally by P4a's per-kind helpers (MG_Impl/Pipe/PipeFill.cpp's
// NotifyAndFree frees whatever the notice does), so PipeSlotPeek never saw a leak.
// What leaked was the SERVER's twin: a framebuffer has NO wire delete opcode at all
// (BRIEF-P4A D-I2), so `object_death` is its only death delivery, and without a
// consumer here the server's record stayed Live for the life of the session.
void OnMagmaFrontendStateObjectDestroyed(MG_Pipe::MGPipeKind kind, Uint64 lifetimeId) {
// Monolith has no record to emit and no server twin to tell.
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
// Raised ON the apply thread: the server role destroying a frontend object it
// created itself (Magma's hidden blit / depth-mipmap resources). The client's
// allocator is a client-thread surface (CONTRACT-P5C §3.1) and EmitObjectDeathRecord
// probes it, so this arm stays silent exactly as Espryt's does.
if (MG_Remote::Server::ServerLoop::OnApplyThread()) return;
(void)MG_Remote::Client::EmitObjectDeathRecord(kind, lifetimeId);
}
const MG_State::GLState::StateObjectDeathOps g_magmaStateObjectDeathOps = {
.OnDestroyed = OnMagmaFrontendStateObjectDestroyed,
};
} // namespace
void InstallStateObjectDeathOps() {
// Installed from BackendObject_DirectVulkan::Initialize(), i.e. step 1 of
// InitServerRoleCommon (ServerLoop::CreateBackend calls Initialize), which is before
// the client session starts and therefore before any frontend object can die.
// Unconditional, exactly as Espryt's ResolveEsprytSlotTablesArm is: only one backend
// is live at a time, and InstallClientWireTables' own "whoever installed first keeps
// the notice" guard is what keeps the remote client from stomping this.
MG_State::GLState::SetStateObjectDeathOps(&g_magmaStateObjectDeathOps);
}
Bool StateObjectDeathOpsInstalled() {
return MG_State::GLState::GetStateObjectDeathOps() == &g_magmaStateObjectDeathOps;
}
#endif // MOBILEGL_BUILD_DISAGGREGATED && MOBILEGL_PIPE_PUSH
Uint64 GetRendererGeneration() {
return g_rendererGeneration.load(std::memory_order_acquire);
}
@@ -64,6 +131,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
namespace {
#if !MOBILEGL_BUILD_DISAGGREGATED
// P7 wave 2 package C, OQ-8: THE PULL BUILD'S HALF, VERBATIM. A disaggregated build
// answers GetShaderStorageBlock{Index,Binding} out of LinkArtifacts::storageBlocks
// instead (see those functions below), which retires this whole cache - the map, the
// SPIRV-Reflect rebuild, the teardown clear and the rehash hazard. It cannot be
// retired HERE because the archive member it reads is itself disaggregated-only, for
// G1: one more Vector in LinkArtifacts moves the pull build's .text, which is pinned
// byte-for-byte at 0xa52203.
//
// NOTHING OUTSIDE THIS FILE READS ANY OF IT (measured: BufferVariableResource,
// bufferVariables, activeVariables and StorageBlockResource have no reference anywhere
// else in the tree, and `dataSize` is written and never read - GL_BUFFER_VARIABLE and
// GL_ACTIVE_VARIABLES are answered by MG_Impl/GLImpl/Program/ProgramInterface.cpp out
// of glslang's own reflection, which is a different index space). So the two arms
// below differ in exactly the two functions their callers use, and in nothing else.
struct BufferVariableResource {
String name;
GLuint blockIndex = 0;
@@ -95,6 +177,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Vector<StorageBlockResource> storageBlocks;
Vector<BufferVariableResource> bufferVariables;
};
#endif // !MOBILEGL_BUILD_DISAGGREGATED
struct DrawElementsIndirectCommand {
Uint32 count = 0;
@@ -111,12 +194,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint32 baseInstance = 0;
};
#if !MOBILEGL_BUILD_DISAGGREGATED
// Keyed by GL program name so the freed-name reuse in IndexGenerator bounds the
// map at the peak-simultaneous-program high-water mark; each slot's ownership is
// checked against the program's lifetime id before it is served (see
// GetProgramResourceCache). Cleared wholesale at EGL teardown via
// ClearProgramResourceCaches.
UnorderedMap<GLuint, ProgramResourceCache> g_programResourceCaches;
#endif
void ClearReadPixelsOutput(GLsizei width, GLsizei height, GLenum format, GLenum type, void* pixels) {
if (!pixels || width <= 0 || height <= 0) {
@@ -131,6 +216,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
#if !MOBILEGL_BUILD_DISAGGREGATED
String NormalizeDescriptorName(const SpvReflectDescriptorBinding& binding) {
const char* rawName = binding.name;
if (binding.type_description != nullptr && binding.type_description->type_name != nullptr) {
@@ -281,18 +367,39 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
return cache;
}
#endif // !MOBILEGL_BUILD_DISAGGREGATED
#if MOBILEGL_BUILD_DISAGGREGATED
// The verb's handles identify server stores. Readback orders GPU-produced
// command/count bytes before CPU tier expansion; never inspect a client binding.
// The verb's handles identify server stores; never inspect a client binding.
//
// P8-D: THE GPU READS THE WORDS (VulkanRenderer::DrawWireIndirectNative, WireDraw.inc), as
// on the monolith arm. The CPU expansion below is kept for the one shape the monolith arm
// also expands - a COUNT form the device cannot issue natively (no VK_KHR_draw_indirect_count,
// or maxdrawcount > 1 without multiDrawIndirect: VulkanRenderer.cpp's
// MultiDrawElementsIndirectCount) - where ReadWireBuffer orders GPU-produced command/count
// bytes before the expansion.
void DrawWireIndirect(GLenum mode, GLenum type, const void* indirect, GLsizei drawcount,
GLsizei stride, Bool indexed, Bool counted = false, GLintptr countOffset = 0) {
if (drawcount <= 0) return;
if (drawcount <= 0 || countOffset < 0) return;
if (pVulkanRenderer->DrawWireIndirectNative(mode, type, reinterpret_cast<Uint64>(indirect), drawcount,
stride, indexed, counted, static_cast<Uint64>(countOffset)))
return;
auto& buffers = pVulkanRenderer->GetWireBufferManager();
const auto& state = MG_Pipe::MGPipeApplier();
// P8-D: this call reads its words on the CPU; the waits those reads take are its own.
struct ExpansionCount {
VkBufferManager& buffers;
const Uint64 waitsBefore = buffers.GetWireHostWaitCount();
~ExpansionCount() {
if (!MG_Util::PipeStats::Enabled()) return;
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireIndirectCpuExpansions, 1);
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireHostWaitsIndirect,
buffers.GetWireHostWaitCount() - waitsBefore);
}
} expansionCount{buffers};
if (counted) {
Uint32 count = 0;
if (countOffset < 0 || !buffers.ReadWireBuffer(state.VerbIndirectParameterBuffer,
if (!buffers.ReadWireBuffer(state.VerbIndirectParameterBuffer,
static_cast<Uint64>(countOffset), sizeof(count), &count)) return;
drawcount = static_cast<GLsizei>(std::min<Uint32>(count, static_cast<Uint32>(drawcount)));
if (!drawcount) return;
@@ -381,6 +488,67 @@ namespace MobileGL::MG_Backend::DirectVulkan {
} // namespace
#if MOBILEGL_BUILD_DISAGGREGATED
// P7 wave 2 package C, OQ-8 (CONTRACT-P7 §5.3): THE MONOLITH CONSUMER READS THE ARCHIVE
// TOO, so one published list serves both of this backend's arms.
//
// THE TWO ARMS BELOW ARE NOT TWO ANSWERS. ProgramTest's
// TheArchivesStorageBlockOrderIsTheOneSpirvReflectProduces runs the OLD arm's algorithm -
// SPIRV-Reflect over the real modules, sorted and deduplicated exactly as
// GetProgramResourceCache does it - over a multi-stage program with an arrayed SSBO and an
// atomic-counter block, and asserts element-for-element equality with what the archive
// published. That equivalence is what makes an #if here a build-time SELECTION rather than
// a behavioural fork, and it is the reason the case exists.
//
// WHY AN #if AT ALL, AND WHAT IT COSTS. LinkArtifacts::storageBlocks is
// MOBILEGL_BUILD_DISAGGREGATED-only, because one more Vector member changes the struct's
// size, its implicit destructor and its move constructor in the PULL build whose .text G1
// pins byte-for-byte (0xa52203). So the pull build keeps `g_programResourceCaches`, the
// SPIRV-Reflect rebuild and `ClearProgramResourceCaches`, verbatim; only this build drops
// them - and with them the rehash hazard the ordering comment below the block-binding
// setter documents. Retiring the pull half is the day the member stops needing its guard,
// which is a G1/P13 decision and not a wave-2 one; recorded in notes/p7/magma-c.md.
void ClearProgramResourceCaches() {
// Nothing to clear: there is no cache in this build. Kept as an entry point so
// BackendObject_DirectVulkan's two EGL-teardown call sites stay statement for
// statement what they are in the pull build.
}
GLuint GetShaderStorageBlockIndex(const MG_State::GLState::ProgramObject& program, const String& name) {
const auto& blocks = program.GetLinkReflection().storageBlocks;
const auto find = [&blocks](const String& key) {
return std::find_if(blocks.begin(), blocks.end(),
[&](const MG_State::GLState::StorageBlockReflection& block) { return block.name == key; });
};
auto it = find(name);
if (it == blocks.end()) {
// Archive names are normalised, so an arrayed block that GL enumerates per element
// - "B[0]", "B[1]" - is one entry here, spelled "B". Same second chance the cache
// gave, for the same callers.
const auto bracket = name.rfind('[');
if (bracket == String::npos || name.empty() || name.back() != ']') return GL_INVALID_INDEX;
it = find(name.substr(0, bracket));
if (it == blocks.end()) return GL_INVALID_INDEX;
}
return static_cast<GLuint>(std::distance(blocks.begin(), it));
}
GLuint GetShaderStorageBlockBinding(const MG_State::GLState::ProgramObject& program, GLuint blockIndex) {
const auto& blocks = program.GetLinkReflection().storageBlocks;
if (blockIndex >= blocks.size()) {
return 0;
}
// THE OVERRIDE IS APPLIED ON READ, where the cache used to apply it on rebuild and
// patch it in place on rebind. Same answer by a shorter route: the program is the
// authoritative record of a rebound block (it is what GL_BUFFER_BINDING reports), the
// archive carries the DECLARED binding only, and asking the program here means a
// rebind can no longer be lost to a cache rebuild that happened to race a state-version
// bump.
const Int rebound = program.GetShaderStorageBlockBindingOverride(blocks[blockIndex].name);
if (rebound >= 0) return static_cast<GLuint>(rebound);
return blocks[blockIndex].binding;
}
#else
void ClearProgramResourceCaches() {
// Called from EGL teardown while the backend's m_eglStateMutex is held; GL
// calls are serialized in this codebase (contexts migrate threads but never
@@ -416,6 +584,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
return cache.storageBlocks[blockIndex].binding;
}
#endif
void ClearBufferfi(GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil) {
MOBILEGL_ASSERT(pVulkanRenderer, "DirectVulkan::ClearBufferfi called with null VulkanRenderer");
@@ -768,12 +937,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void DispatchComputeIndirect(GLintptr indirect) {
#if MOBILEGL_BUILD_DISAGGREGATED
// P8-SV: THE GPU READS THE GROUP COUNTS on the wire arm too - VulkanRenderer's
// DispatchComputeIndirect takes its transport branch into DispatchWireComputeIndirect
// (WireDraw.inc), which issues vkCmdDispatchIndirect from the verb's store. The CPU read
// that was here (ReadWireBuffer, a whole-GPU wait once a shader had written the store) is gone.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
Uint32 groups[3]{};
if (indirect >= 0 && pVulkanRenderer->GetWireBufferManager().ReadWireBuffer(
MG_Pipe::MGPipeApplier().VerbDispatchIndirectBuffer,
static_cast<Uint64>(indirect), sizeof(groups), groups))
pVulkanRenderer->DispatchCompute(groups[0], groups[1], groups[2]);
pVulkanRenderer->DispatchComputeIndirect(indirect);
return;
}
#endif
@@ -882,6 +1051,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MakeUnique<GenericErrorInfo>("DirectVulkan", __func__, "Shader storage binding is out of range."));
return;
}
#if MOBILEGL_BUILD_DISAGGREGATED
// P7 wave 2 package C, OQ-8: NOTHING LEFT TO PATCH, and the hazard goes with it.
//
// The frontend has already recorded the new binding on the program, and that record is
// now what GetShaderStorageBlockBinding reads (it asks
// GetShaderStorageBlockBindingOverride on every call, over an IMMUTABLE archive list).
// What stood here was the other half of a mutable cache: resolve the index, take a
// reference into g_programResourceCaches and patch `binding` in place so an
// already-built entry need not be thrown away.
//
// THE HAZARD THAT CAME WITH IT IS GONE TOO, and it is worth naming because it was a
// real crash rather than a theoretical one: GetShaderStorageBlockIndex re-entered
// GetProgramResourceCache, which indexes an OPEN-ADDRESSED map and can therefore
// insert, and a rehash moves entries - so a reference taken before that call dangled.
// Binding one program's storage block while another program's entry was still absent
// from the cache was a reproducible segfault (ProgramPipelineScenario's two
// storage-block cases, in one process). The ordering above was the fix; having no
// cache is the retirement. The pull build below keeps both.
(void)storageBlockBinding;
}
#else
// The frontend already validated that the name denotes an active block, and has
// already recorded the new binding on the program - which is what reseeds this cache
// whenever it is rebuilt. Writing the entry here as well keeps an ALREADY-BUILT cache
@@ -900,6 +1090,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (blockIndex >= cache.storageBlocks.size()) return;
cache.storageBlocks[blockIndex].binding = storageBlockBinding;
}
#endif
void ReadPixels(GLint x, GLint y, GLsizei width, GLsizei height, GLenum format, GLenum type, void* pixels) {
MOBILEGL_ASSERT(pVulkanRenderer, "DirectVulkan::ReadPixels called with null VulkanRenderer");
MOBILEGL_ASSERT(MGB_CTX_LIVE, "DirectVulkan::ReadPixels called with null GL context");
@@ -1570,4 +1761,52 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MG_Util::PipeStats::OnPresent();
}
}
namespace {
Optional<Int> g_requestedSwapInterval;
} // namespace
void SetSwapInterval(Int interval) {
g_requestedSwapInterval = interval;
if (pVulkanRenderer) {
pVulkanRenderer->SetSwapInterval(interval);
}
}
Optional<Int> GetRequestedSwapInterval() {
return g_requestedSwapInterval;
}
#if MOBILEGL_BUILD_DISAGGREGATED
void ForgetRequestedSwapInterval() {
g_requestedSwapInterval.reset();
}
#endif
DescriptorPoolCensus GetDescriptorPoolCensus() {
#if MOBILEGL_BUILD_DISAGGREGATED
// P11 M2: UNDER A TRANSPORT THE POOLS ARE THE SERVER RENDERER'S, and its apply thread
// grows, rewinds and trims them while the test thread would be reading - a census taken
// after a swap the client ran ahead of, or mid-burst, walks a set cache BeginFrame may be
// clearing. Inproc, the server is a thread of this process: the read is posted to it as
// the test probe, which runs between drain batches, so it sees every record published
// before it and none half-applied - without waiting on the GPU, which is the point of the
// trim case (it gives a burst back with other slots' work in flight). Spawn and tcp keep
// their pools in another process; the client has no renderer and answers unavailable.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith &&
!MG_Remote::Server::ServerLoop::OnApplyThread()) {
if (MG_Config::Transport != MG_Config::TransportMode::InProcess) return DescriptorPoolCensus{};
DescriptorPoolCensus census{};
const MobileGLResult result = MG_Remote::Server::ServerLoopInstance().RunProbeOnApplyThreadForTesting(
+[](void* user) -> MobileGLResult {
*static_cast<DescriptorPoolCensus*>(user) =
pVulkanRenderer ? pVulkanRenderer->GetDescriptorPoolCensus() : DescriptorPoolCensus{};
return MOBILEGL_OK;
},
&census);
return result == MOBILEGL_OK ? census : DescriptorPoolCensus{};
}
#endif
return pVulkanRenderer ? pVulkanRenderer->GetDescriptorPoolCensus() : DescriptorPoolCensus{};
}
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -23,6 +23,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint64 GetRendererGeneration();
void BumpRendererGeneration();
#if MOBILEGL_BUILD_DISAGGREGATED && MOBILEGL_PIPE_PUSH
// P7 wave 2 package C (CONTRACT-P7 §5.5). Magma's StateObjectDeathOps table, installed
// from BackendObject_DirectVulkan::Initialize() so that GetStateObjectDeathOps() is
// non-null on every Magma role that owns a backend. See the definition for what the one
// arm does and for the inproc gap it closes. Push-only and disaggregated-only: a pull
// build declares no notice at all (StateObjectDeathNotice.h) and has nothing to emit.
void InstallStateObjectDeathOps();
// True when the notice currently installed is this backend's. Read by MG_Test so the
// install can be asserted without a death having to happen first.
Bool StateObjectDeathOpsInstalled();
#endif
// Drops every cached program-resource reflection entry (CPU-side strings/vectors
// only, no Vulkan handles). Called at EGL teardown next to the renderer reset;
// safe because GL calls are serialized in this codebase, and any still-live
@@ -135,4 +147,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// only exist as vkCmdWriteTimestamp results); the frontend falls back.
Int64 GetGpuTimestampNs();
void Present();
// eglSwapInterval. Remembered across renderers: one created for a later window surface
// starts from the app's last request (GetRequestedSwapInterval) instead of the default.
void SetSwapInterval(Int interval);
Optional<Int> GetRequestedSwapInterval();
#if MOBILEGL_BUILD_DISAGGREGATED
// The swap interval an ended server session asked for is not the next one's, the same rule as
// DirectGLES::ForgetRequestedSwapInterval: under a transport ~BackendObject_DirectVulkan forgets
// it, so the next session starts at the default until it asks.
void ForgetRequestedSwapInterval();
#endif
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -307,8 +307,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (result != VK_SUCCESS) {
return result;
}
// The slot's fence has been waited: every command buffer this slot
// submitted (including mid-frame flushes) has finished executing.
// Under disaggregation VulkanRenderer waits every registered fence
// through this slot's last submit before calling here. That aggregate
// proof covers mid-frame pooled-fence flushes too; the slot fence alone
// would not. Initialize's first acquire has no prior submissions.
FreeRetiredCommandBuffers(frame);
result = vkAcquireNextImageKHR(device, swapchain, timeout, frame.imageAvailableSemaphore, acquireFence,
@@ -166,9 +166,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// WHAT IT DOES NOT COVER, AND WHY NO REPRODUCER OF THIS SHAPE CAN [fix-aba review v1,
// MAJOR 1]. It does NOT exercise the GENERATION half of {slot, gen}:
//
// * this mint has no death notification - nothing in MG_Backend/DirectVulkan consumes
// NotifyStateObjectDestroyed - so a slot returns to the free list only through
// OnFrameBoundary's age sweep (kSweepInterval 256, kRetireAgeBoundaries 1024, below);
// * THIS MINT still has no death notification, and that is now a narrower statement than
// it was. P7 wave 2 package C gives DirectVulkan a StateObjectDeathOps table
// (DirectVulkan.cpp's g_magmaStateObjectDeathOps, CONTRACT-P7 §5.5), so
// NotifyStateObjectDestroyed DOES have a consumer on this backend - but that consumer's
// one job is to emit the `object_death` RECORD so the server drops its twin, and it
// never touches these per-renderer identity tables: they are the SERVER-side renderer's
// own, the notice is raised on the CLIENT thread, and an Acquire/retire from the wrong
// side is precisely the cross-role read CONTRACT-P5C §3.1 forbids. So a slot here still
// returns to the free list only through OnFrameBoundary's age sweep (kSweepInterval
// 256, kRetireAgeBoundaries 1024, below), and everything the two bullets below derive
// from that is unchanged;
// * HandleRecycleScenario issues five frame boundaries, so the free list is empty when
// the replacement VAO acquires and it gets a BRAND-NEW slot at Gen 1 (measured:
// redVao slot=2 gen=1, greenVao slot=3 gen=1). The knob-off FRESH verdict there is
@@ -7,7 +7,11 @@
#include <MG_Pipe/PipeApply.h>
#include <MG_State/GLState/ProgramState/ProgramArtifactsCodec.h>
#include "../DirectVulkanResourceState.h"
#include <spirv_reflect.h>
// P7 wave 2 package C, OQ-8: spirv_reflect.h is GONE from this header, and its absence is the
// gate. Everything this class needed it for - the storage-block index space - now travels in
// the archive (LinkArtifacts::storageBlocks), so the SPIR-V reflector is off the Magma wire
// draw path entirely. A future edit that re-adds the include is re-adding a per-draw reflect,
// because a MagmaProgramSource is constructed once per draw and can memoise nothing.
#endif
namespace MobileGL::MG_Backend::DirectVulkan {
@@ -153,30 +157,49 @@ namespace MobileGL::MG_Backend::DirectVulkan {
GLenum GetTransformFeedbackBufferMode() const {
return IsWire() ? Link().xfbBufferMode : m_frontend->GetTransformFeedbackBufferMode();
}
// P7 wave 2 package C, OQ-8 (CONTRACT-P7 §5.3): READ OUT OF THE ARCHIVE.
//
// What stood here re-ran spvReflectCreateShaderModule over EVERY stage module of the
// program, and it did so PER DRAW: a MagmaProgramSource is a borrowed view that
// WireDraw.inc's WireProgramSource constructs afresh for each wire draw, so the lazy
// `m_storageBlocksReady` latch could never outlive one. The list it rebuilt is a pure
// function of the linked program, so it is now built once at link time and travels in
// the archive (LinkArtifacts::storageBlocks), which takes the SPIR-V reflector off the
// wire draw path entirely - this header no longer includes spirv_reflect.h at all.
Uint GetShaderStorageBlockIndex(const String& name) const {
if (!IsWire()) return DirectVulkan::GetShaderStorageBlockIndex(*m_frontend, name);
EnsureStorageBlocks();
String key = name;
auto find = [&] {
return std::find_if(m_storageBlocks.begin(), m_storageBlocks.end(),
[&](const StorageBlock& block) { return block.Name == key; });
const auto& blocks = Link().storageBlocks;
// The archive carries NORMALISED names (the array subscript stripped), because
// that is the spelling SPIRV-Reflect's type_name gives and the spelling
// ProgramFactory::ReflectLayout looks up. A caller that still holds a GL-style
// "Foo[2]" gets the same second chance it always had.
const auto find = [&blocks](const String& key) {
return std::find_if(blocks.begin(), blocks.end(),
[&](const MG_State::GLState::StorageBlockReflection& block) {
return block.name == key;
});
};
auto found = find();
if (found == m_storageBlocks.end() && !name.empty() && name.back() == ']') {
auto found = find(name);
if (found == blocks.end() && !name.empty() && name.back() == ']') {
const auto bracket = name.rfind('[');
if (bracket != String::npos) { key = name.substr(0, bracket); found = find(); }
if (bracket != String::npos) found = find(name.substr(0, bracket));
}
return found == m_storageBlocks.end() ? GL_INVALID_INDEX : static_cast<Uint>(found - m_storageBlocks.begin());
return found == blocks.end() ? GL_INVALID_INDEX
: static_cast<Uint>(found - blocks.begin());
}
Uint GetShaderStorageBlockBinding(Uint index) const {
if (!IsWire()) return DirectVulkan::GetShaderStorageBlockBinding(*m_frontend, index);
EnsureStorageBlocks();
if (index >= m_storageBlocks.size()) return 0;
const auto& block = m_storageBlocks[index];
const auto& blocks = Link().storageBlocks;
if (index >= blocks.size()) return 0;
const auto& block = blocks[index];
// THE OVERRIDE STILL WINS, and it still comes off the RECORD rather than the
// archive: glShaderStorageBlockBinding is a post-link mutable and the archive
// carries link-time defaults only (rule on MagmaProgramSource, and the reason
// StorageBlockReflection::binding is documented as the DECLARED binding).
for (const auto& override : m_record->StorageOverrides) {
if (override.Name == block.Name) return override.Binding;
if (override.Name == block.name) return override.Binding;
}
return block.Binding;
return block.binding;
}
private:
@@ -187,48 +210,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return tIndex >= 0 && static_cast<SizeT>(tIndex) < Link().blockReflection.size()
? &Link().blockReflection[tIndex] : nullptr;
}
struct StorageBlock { String Name; Uint Binding; };
void EnsureStorageBlocks() const {
if (m_storageBlocksReady) return;
m_storageBlocksReady = true;
// Match the legacy resource-query index order: stage order, then
// normalized descriptor name/binding, deduplicated across stages.
for (const auto& spirv : GetGeneratedSpirv()) {
if (spirv.empty()) continue;
SpvReflectShaderModule module{};
if (spvReflectCreateShaderModule(spirv.size() * sizeof(Uint), spirv.data(), &module) != SPV_REFLECT_RESULT_SUCCESS) continue;
Uint32 count = 0;
if (spvReflectEnumerateDescriptorBindings(&module, &count, nullptr) == SPV_REFLECT_RESULT_SUCCESS) {
Vector<SpvReflectDescriptorBinding*> bindings(count);
if (spvReflectEnumerateDescriptorBindings(&module, &count, bindings.data()) == SPV_REFLECT_RESULT_SUCCESS) {
auto name = [](const SpvReflectDescriptorBinding* binding) {
const char* raw = binding->name;
if (binding->type_description && binding->type_description->type_name) raw = binding->type_description->type_name;
String result = raw ? raw : "";
const auto suffix = result.find("[0]");
return suffix == String::npos ? result : result.substr(0, suffix);
};
std::sort(bindings.begin(), bindings.end(), [&](const auto* a, const auto* b) {
const String an = a ? name(a) : String(), bn = b ? name(b) : String();
return an != bn ? an < bn : (a && b && a->binding < b->binding);
});
for (const auto* binding : bindings) {
if (!binding || binding->descriptor_type != SPV_REFLECT_DESCRIPTOR_TYPE_STORAGE_BUFFER) continue;
const String key = name(binding);
if (key.empty() || std::any_of(m_storageBlocks.begin(), m_storageBlocks.end(),
[&](const StorageBlock& block) { return block.Name == key; })) continue;
m_storageBlocks.push_back({key, binding->binding});
}
}
}
spvReflectDestroyShaderModule(&module);
}
}
const Program* m_frontend = nullptr;
MG_Pipe::MGPipeHandle m_handle = MG_Pipe::kMGPipeNullHandle;
const MG_Pipe::MGPipeShaderCsoRecord* m_record = nullptr;
mutable Bool m_storageBlocksReady = false;
mutable Vector<StorageBlock> m_storageBlocks;
// P7 OQ-8: the two mutables that memoised the per-draw reflect are gone with it. They
// were `mutable` precisely because this class is a borrowed const view - and that is
// also why they never helped: a view lives for one draw, so the latch was cold every
// time it was read.
};
#else
using MagmaProgramSource = MG_State::GLState::ProgramObject;
@@ -0,0 +1,41 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/SubmitFencePrefix.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <cstdint>
#include <vector>
namespace MobileGL::MG_Backend::DirectVulkan {
// Collect every fence in the ordered submission prefix. Waiting on only the last matching
// frame-serial record is insufficient: the same serial may have an earlier in-flight submit.
template <typename Records, typename Fence>
bool CollectSubmitFencePrefix(const Records& records, std::uint64_t submitIndex,
Fence nullFence, std::vector<Fence>& fences) {
fences.clear();
bool foundTarget = false;
for (const auto& record : records) {
if (record.submitIndex > submitIndex) {
break;
}
if (record.fence == nullFence) {
fences.clear();
return false;
}
fences.push_back(record.fence);
foundTarget = foundTarget || record.submitIndex == submitIndex;
}
if (!foundTarget) {
fences.clear();
return false;
}
return true;
}
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -8,6 +8,8 @@
#include "SwapchainObject.h"
#include <span>
#include "MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h"
#include "MG_State/GLState/TextureState/TextureObject2D.h"
@@ -140,12 +142,26 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
VkPresentModeKHR SwapchainObject::ChooseSwapchainPresentMode(
const Vector<VkPresentModeKHR>& availablePresentModes) {
for (auto desiredPresentMode : s_desiredPresentModes) {
for (const auto& presentMode : availablePresentModes) {
if (presentMode == desiredPresentMode) {
return presentMode;
}
const Vector<VkPresentModeKHR>& availablePresentModes, Optional<Int> swapInterval) {
static constexpr VkPresentModeKHR s_unthrottledPresentModes[]{
VK_PRESENT_MODE_IMMEDIATE_KHR, VK_PRESENT_MODE_MAILBOX_KHR, VK_PRESENT_MODE_FIFO_KHR};
static constexpr VkPresentModeKHR s_vsyncPresentModes[]{VK_PRESENT_MODE_FIFO_KHR};
static constexpr VkPresentModeKHR s_adaptiveVsyncPresentModes[]{VK_PRESENT_MODE_FIFO_RELAXED_KHR,
VK_PRESENT_MODE_FIFO_KHR};
std::span<const VkPresentModeKHR> ranked = s_desiredPresentModes;
if (swapInterval) {
if (*swapInterval == 0) {
ranked = s_unthrottledPresentModes;
} else if (*swapInterval > 0) {
ranked = s_vsyncPresentModes;
} else {
ranked = s_adaptiveVsyncPresentModes;
}
}
for (auto desiredPresentMode : ranked) {
if (std::find(availablePresentModes.begin(), availablePresentModes.end(), desiredPresentMode) !=
availablePresentModes.end()) {
return desiredPresentMode;
}
}
@@ -153,9 +169,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return availablePresentModes[0];
}
const char* SwapchainObject::GetPresentModeName(VkPresentModeKHR presentMode) {
return string_VkPresentModeKHR(presentMode);
}
void SwapchainObject::Create(VkDevice device, VkPhysicalDevice physicalDevice, VkSurfaceKHR surface,
Uint32 graphicsQueueFamily, Uint32 presentQueueFamily, Uint32 minImageCountHint,
VkExtent2D desiredExtent) {
VkExtent2D desiredExtent, Optional<Int> swapInterval) {
const auto swapchainCapabilities = GetSwapchainCapabilities(physicalDevice, surface);
MOBILEGL_ASSERT(swapchainCapabilities.IsComplete(),
"SwapchainObject::Create failed: incomplete swapchain capabilities");
@@ -174,8 +194,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MGLOG_D(" %s", string_VkPresentModeKHR(pm));
}
const auto presentMode = ChooseSwapchainPresentMode(swapchainCapabilities.presentModes);
const auto presentMode = ChooseSwapchainPresentMode(swapchainCapabilities.presentModes, swapInterval);
MGLOG_I("Picked present mode: %s", string_VkPresentModeKHR(presentMode));
m_presentMode = presentMode;
m_supportedPresentModes = swapchainCapabilities.presentModes;
const auto& swapchainCaps = swapchainCapabilities.capabilities;
Uint32 targetImageCount = std::max<Uint32>(minImageCountHint, swapchainCaps.minImageCount);
@@ -26,13 +26,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
static SwapchainCapabilities GetSwapchainCapabilities(VkPhysicalDevice physicalDevice, VkSurfaceKHR surface);
static VkSurfaceFormatKHR ChooseSwapchainSurfaceFormat(const Vector<VkSurfaceFormatKHR>& availableFormats);
static VkPresentModeKHR ChooseSwapchainPresentMode(const Vector<VkPresentModeKHR>& availablePresentModes);
// swapInterval is the app's eglSwapInterval, empty until it asks (s_desiredPresentModes
// ranks the modes then). 0 wants IMMEDIATE, else MAILBOX, else FIFO; >= 1 wants FIFO,
// which cannot skip vblanks, so 2 and up pace like 1; a negative interval (adaptive
// vsync) wants FIFO_RELAXED, else FIFO.
static VkPresentModeKHR ChooseSwapchainPresentMode(const Vector<VkPresentModeKHR>& availablePresentModes,
Optional<Int> swapInterval = Nullopt);
static const char* GetPresentModeName(VkPresentModeKHR presentMode);
void Create(VkDevice device, VkPhysicalDevice physicalDevice, VkSurfaceKHR surface, Uint32 graphicsQueueFamily,
Uint32 presentQueueFamily, Uint32 minImageCountHint, VkExtent2D desiredExtent);
Uint32 presentQueueFamily, Uint32 minImageCountHint, VkExtent2D desiredExtent,
Optional<Int> swapInterval);
void Shutdown(VkDevice device);
VkSwapchainKHR GetHandle() const { return m_swapchain; }
VkPresentModeKHR GetPresentMode() const { return m_presentMode; }
// The surface's present modes as queried when this swapchain was created.
const Vector<VkPresentModeKHR>& GetSupportedPresentModes() const { return m_supportedPresentModes; }
const VkSurfaceFormatKHR& GetSurfaceFormat() const { return m_surfaceFormat; }
VkExtent2D GetExtent() const { return m_extent; }
// Surface-space extent (before the pre-rotation quarter-turn swap) this swapchain was
@@ -79,6 +89,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
};
VkSwapchainKHR m_swapchain = VK_NULL_HANDLE;
VkPresentModeKHR m_presentMode = VK_PRESENT_MODE_FIFO_KHR;
Vector<VkPresentModeKHR> m_supportedPresentModes;
VkSurfaceFormatKHR m_surfaceFormat{};
VkExtent2D m_extent{};
VkExtent2D m_surfaceExtent{};
@@ -31,6 +31,10 @@
#include <Config.h>
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Pipe/PipeApply.h>
// P7 wave 0: the seam WireDescriptorFatal dies through. See MG_Pipe/PipeSessionFail.h.
#include <MG_Pipe/PipeSessionFail.h>
// P7 wave 2 package B3: rule I's tally for this file's silent wire-draw exits.
#include "WireDeclineTally.h"
#endif
#include <vulkan/utility/vk_format_utils.h>
#include <algorithm>
@@ -65,9 +69,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return hash;
}
// P7 wave 0, the descriptor half of WireFramebuffer.inc's MagmaWireFatal - same message,
// same seam, same reason. Four of the fifteen P7-marked refusals die here.
[[noreturn]] static void WireDescriptorFatal(const char* detail) {
MGLOG_F("MGPipe: Fatal{UnmigratedVerb, \"Magma:%s\"}", detail);
std::abort();
MG_Pipe::MGPipeSessionFail(MG_Pipe::MGPipeFatalFamily::UnmigratedVerb,
"MGPipe: Fatal{UnmigratedVerb, \"Magma:%s\"}", detail);
}
static Bool ResolveWireRange(Uint64 offset, Uint64 declaredSize, VkDeviceSize bufferSize,
@@ -93,6 +99,49 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
// GL ignores Layer/Layered for a target with no layers. In particular image1D stays a 1D
// view, and an ignored nonzero Layer must not address another slice.
static Bool WireTargetHasLayers(MG_Pipe::MGPipeResourceTarget target) {
using T = MG_Pipe::MGPipeResourceTarget;
return target == T::Tex1DArray || target == T::Tex2DArray || target == T::Tex2DMSArray ||
target == T::Tex3D || target == T::TexCube || target == T::TexCubeArray;
}
// GL 4.6 core 8.26: an access through an image unit whose texture does not HAVE the
// (level, layer) the unit names is INVALID. A load returns zero, a store does nothing, an
// atomic updates nothing - and none of it is an error, neither at bind time (the bind only
// checks the level's and the layer's signs) nor at the draw. It is the observable an empty
// unit has, so the answer is the empty unit's (ResolveWirePlaceholderImage): a NULL storage
// descriptor where the device enables VK_EXT_robustness2 nullDescriptor, else a storage
// placeholder private to the unit (per shape), cleared before each use, of the dimensionality
// the SHADER declared.
//
// KHR-GL46.shader_image_load_store.incomplete_textures is the shape that found it: level 2
// of a texture that defines only level 0 (MAX_LEVEL 7, a mipmapping filter). The same
// question comes up empty for a level past an immutable texture's storage, a layer past an
// array's last slice, a texture that has no storage at all, and a level past a texture
// view's own window even where its owner has one. The record carries exactly what
// glBindImageTexture was given, so it is decided here, by the walk every wire consumer uses
// to find a texel's storage, and from that walk's "not there" answer ONLY: a null for any
// other reason (a dead record, a stale view CSO) stays the descriptor resolve's refusal.
//
// NOT DECIDED HERE, deliberately: a level that IS in storage but lies outside
// [BASE_LEVEL, MAX_LEVEL], or of a texture that is not mipmap-complete. The monolith arm
// binds those as it finds them too, and the P7 gate 5 comparison is against that arm.
static Bool WireShaderImageNamesNoTexel(VkTextureManager& textures, const MG_Pipe::MGPImageView& image) {
const auto& state = MG_Pipe::MGPipeApplier();
if (MG_Pipe::MGPipeHandleIsNull(image.Res) || image.Res.Slot >= state.TextureResources.size()) return false;
const auto& record = state.TextureResources[image.Res.Slot];
if (!record.Live || record.Gen != image.Res.Gen) return false;
const Bool selectsLayer =
WireTargetHasLayers(static_cast<MG_Pipe::MGPipeResourceTarget>(record.Desc.Target)) && !image.Layered;
Uint32 level = image.Level;
Uint32 layer = selectsLayer ? image.Layer : 0;
Bool outsideWindow = false;
(void)textures.ResolveWireTextureStorage(image.Res, level, layer, nullptr, nullptr, &outsideWindow);
return outsideWindow;
}
static VkComponentSwizzle WireSwizzle(Uint8 value) {
switch (static_cast<TextureSwizzleParam>(value)) {
case TextureSwizzleParam::Red: return VK_COMPONENT_SWIZZLE_R;
@@ -118,23 +167,31 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (unit < 0 || static_cast<Uint32>(unit) >=
(storage ? MG_Pipe::kMGPipeMaxImageUnits : MG_Pipe::kMGPipeMaxTextureUnits)) return false;
const auto handle = storage ? state.BoundShaderImages[unit].Res : state.BoundSamplerViews[unit].Texture;
// An EMPTY image unit is the same 8.26 invalid access as the named-but-missing texel
// below, so it takes the same answer: a null storage descriptor where the device has
// one, else the unit-private placeholder (codex closeout finding 2).
if (MG_Pipe::MGPipeHandleIsNull(handle))
return ResolveWirePlaceholderImage(commandBuffer, program, programObj, binding, storage, out);
return ResolveWirePlaceholderImage(commandBuffer, program, programObj, binding, storage, out,
VK_FORMAT_UNDEFINED, static_cast<Uint32>(unit));
if (handle.Slot >= state.TextureResources.size()) WireDescriptorFatal("image-record");
const auto& record = state.TextureResources[handle.Slot];
if (!record.Live || record.Gen != handle.Gen) WireDescriptorFatal("image-record-generation");
// Before the sync: a texture with no storage at all is one of the shapes, and syncing
// it would refuse rather than answer.
if (storage && WireShaderImageNamesNoTexel(*m_textureManager, state.BoundShaderImages[unit])) {
const VkFormat bound = MG_Util::ConvertTextureInternalFormatToVkEnum(
MG_Util::ConvertGLEnumToTextureInternalFormat(state.BoundShaderImages[unit].InternalFormat));
return ResolveWirePlaceholderImage(commandBuffer, program, programObj, binding, storage, out, bound,
static_cast<Uint32>(unit));
}
auto* resource = m_textureManager->SyncTextureResourceByHandle(handle, false, storage);
if (!resource) WireDescriptorFatal("image-resource");
m_textureManager->FlushPendingUploads();
const auto target = static_cast<MG_Pipe::MGPipeResourceTarget>(record.Desc.Target);
using Target = MG_Pipe::MGPipeResourceTarget;
const Bool layerable = target == Target::Tex1DArray || target == Target::Tex2DArray ||
target == Target::Tex2DMSArray || target == Target::Tex3D ||
target == Target::TexCube || target == Target::TexCubeArray;
const Bool layerable = WireTargetHasLayers(target);
Uint32 level = storage ? state.BoundShaderImages[unit].Level : record.Params.BaseLevel;
// GL ignores Layer/Layered for non-layerable targets. In particular image1D
// stays a 1D view, and an ignored nonzero Layer must not address another slice.
// GL ignores Layer/Layered for non-layerable targets (WireTargetHasLayers).
Uint32 layer = storage && layerable && !state.BoundShaderImages[unit].Layered
? state.BoundShaderImages[unit].Layer : 0;
const Uint32 localLevel = level;
@@ -534,6 +591,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_samplerResolveMemoHighWater = 0;
m_setsPerFrame = setsPerFrame;
m_peakDescriptorSetsObserved = 0;
m_trimQuietEpochs = MG_Config::Features.MagmaDescriptorTrimFrames;
// The renderer records its first frame into slot 0 before any BeginFrame runs.
m_epochSlot = 0;
m_epochOpen = true;
m_descriptorSlotTrims = 0;
m_textureManager = textureManager;
m_samplerManager = samplerManager;
@@ -543,6 +605,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
frame.activeDescriptorPoolIndex = 0;
frame.allocatedSetsThisFrame = 0;
frame.peakAllocatedSetsThisFrame = 0;
frame.capacitySets = m_setsPerFrame;
frame.quietEpochs = 0;
frame.descriptorPools.clear();
VkDescriptorPool initialPool = VK_NULL_HANDLE;
@@ -597,12 +661,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
frame.activeDescriptorPoolIndex = 0;
frame.allocatedSetsThisFrame = 0;
frame.peakAllocatedSetsThisFrame = 0;
frame.capacitySets = 0;
frame.quietEpochs = 0;
}
m_frames.clear();
#if MOBILEGL_BUILD_DISAGGREGATED
for (auto& entry : m_wirePlaceholderImages) DestroyWirePlaceholderImage(entry.second);
m_wirePlaceholderImages.clear();
m_wirePrivateStoragePlaceholders = 0;
m_wireInvalidStorageImagesBindNull = false;
#endif
m_epochSlot = 0;
m_epochOpen = false;
m_bufferManager = nullptr;
m_programFactory = nullptr;
@@ -642,6 +712,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
"UniformDescriptorBinder: new descriptor set peak observed=%u (base setsPerFrame=%u, frame=%u, pools=%zu)",
m_peakDescriptorSetsObserved, m_setsPerFrame, frameIndex, frame.descriptorPools.size());
}
// The epoch that just ended is the one of the slot begun last - this slot again on the
// drain path, the previous frame's slot on the Present path. Its peak is still unreset.
if (m_epochOpen && m_epochSlot < m_frames.size()) {
NoteEndedEpochForTrim(m_frames[m_epochSlot].peakAllocatedSetsThisFrame);
}
m_epochSlot = frameIndex;
m_epochOpen = true;
// No pending command buffer references this slot's sets here (see TrimFrameDescriptorPools
// for why that holds at both of BeginFrame's call sites), so a quiet grown slot can be
// handed back whole.
if (m_trimQuietEpochs != 0 && frame.descriptorPools.size() > 1 &&
frame.capacitySets >= kDescriptorTrimMinCapacity && frame.quietEpochs >= m_trimQuietEpochs) {
TrimFrameDescriptorPools(frame, frameIndex);
}
frame.activeDescriptorPoolIndex = 0;
frame.allocatedSetsThisFrame = 0;
frame.peakAllocatedSetsThisFrame = 0;
@@ -650,11 +734,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
// The frame's descriptor sets are recycled above, so last frame's reuse targets
// are gone: start the per-draw descriptor-reuse cache fresh this frame.
for (auto& entry : m_descriptorReuseMemo) {
entry.valid = false;
}
m_fastRebindMemo.valid = false;
m_lastBindValid = false;
InvalidateDescriptorSetMemos();
// Re-fingerprint the bound sampler set fresh this frame so any GL object address
// reuse cannot outlive a single frame (see SamplerResolveMemo). Only the entries a
// resolve has actually written can be valid, so the high-water mark bounds the
@@ -668,6 +748,103 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
#if MOBILEGL_BUILD_DISAGGREGATED
Bool UniformManager::WireDescriptorSetBudgetReached(Uint32 frameIndex) const {
MOBILEGL_ASSERT(frameIndex < m_frames.size(), "WireDescriptorSetBudgetReached invalid frame index");
return m_frames[frameIndex].allocatedSetsThisFrame >= kWireDescriptorSetBudget;
}
SizeT UniformManager::RewindWireDescriptorSets(Uint32 frameIndex) {
MOBILEGL_ASSERT(frameIndex < m_frames.size(), "RewindWireDescriptorSets invalid frame index");
auto& frame = m_frames[frameIndex];
// The caller proved that the last graphics submit using these sets has
// retired and that no unsubmitted command buffer still references them.
// Keep the pools and sets: only the per-layout write cursors rewind.
// Image and buffer views stay alive until the regular frame boundary.
m_peakDescriptorSetsObserved = std::max(m_peakDescriptorSetsObserved, frame.peakAllocatedSetsThisFrame);
SizeT cachedSets = 0;
for (auto& entry : frame.descriptorSetCacheByLayout) {
cachedSets += entry.second.sets.size();
entry.second.cursor = 0;
}
frame.allocatedSetsThisFrame = 0;
// The PEAK is not reset: a rewind ends a window of the epoch, not the epoch. The trim scores
// an epoch by the most sets the slot held at once in it (NoteEndedEpochForTrim at the next
// BeginFrame), and on this arm that is the largest window - resetting here would score a
// burst frame by its last window alone and trim a slot that re-grows every frame.
InvalidateDescriptorSetMemos();
const Uint32 touchedBindings =
std::min<Uint32>(m_samplerResolveMemoHighWater, static_cast<Uint32>(m_samplerResolveMemo.size()));
for (Uint32 binding = 0; binding < touchedBindings; ++binding) {
m_samplerResolveMemo[binding].valid = false;
m_samplerResolveMemo[binding].infoValid = false;
}
return cachedSets;
}
#endif
void UniformManager::InvalidateDescriptorSetMemos() {
// Every VkDescriptorSet handle held outside the per-layout caches - in this backend they all
// live in this class. A trim can hand a destroyed set's handle value to a fresh allocation,
// so the bind-dedup shadow must drop too, not only the two reuse memos.
for (auto& entry : m_descriptorReuseMemo) {
entry.valid = false;
}
m_fastRebindMemo.valid = false;
m_lastBindValid = false;
}
void UniformManager::NoteEndedEpochForTrim(Uint32 epochPeakSets) {
for (auto& slot : m_frames) {
if (static_cast<Uint64>(epochPeakSets) * kDescriptorTrimHeadroom > slot.capacitySets) {
slot.quietEpochs = 0;
} else if (slot.quietEpochs < std::numeric_limits<Uint32>::max()) {
++slot.quietEpochs;
}
}
}
void UniformManager::TrimFrameDescriptorPools(FrameResources& frame, Uint32 frameIndex) {
const SizeT poolsBefore = frame.descriptorPools.size();
const Uint64 capacityBefore = frame.capacitySets;
// Holders first: the per-layout cache and every memo that may name one of these sets.
frame.descriptorSetCacheByLayout.clear();
InvalidateDescriptorSetMemos();
for (SizeT i = 1; i < frame.descriptorPools.size(); ++i) {
if (frame.descriptorPools[i].handle != VK_NULL_HANDLE) {
vkDestroyDescriptorPool(m_device, frame.descriptorPools[i].handle, nullptr);
}
}
frame.descriptorPools.resize(1);
auto& base = frame.descriptorPools[0];
// Frees every set allocated from it; the spec defines no failure for vkResetDescriptorPool.
vkResetDescriptorPool(m_device, base.handle, 0);
base.allocatedSets = 0;
frame.activeDescriptorPoolIndex = 0;
frame.capacitySets = base.maxSets;
frame.quietEpochs = 0;
++m_descriptorSlotTrims;
MGLOG_D("UniformDescriptorBinder: frame %u trimmed its descriptor pools (%zu pools / %llu sets -> 1 / %u)",
frameIndex, poolsBefore, static_cast<unsigned long long>(capacityBefore), base.maxSets);
}
DescriptorPoolCensus UniformManager::GetDescriptorPoolCensus() const {
DescriptorPoolCensus census;
census.available = true;
census.frameSlots = static_cast<uint32_t>(m_frames.size());
for (const auto& frame : m_frames) {
const auto poolCount = static_cast<uint32_t>(frame.descriptorPools.size());
census.pools += poolCount;
census.maxPoolsInOneSlot = std::max(census.maxPoolsInOneSlot, poolCount);
census.capacitySets += frame.capacitySets;
for (const auto& cacheEntryPair : frame.descriptorSetCacheByLayout) {
census.cachedSets += cacheEntryPair.second.sets.size();
}
}
census.slotTrims = m_descriptorSlotTrims;
return census;
}
void UniformManager::OnDescriptorSetLayoutDestroyed(VkDescriptorSetLayout descriptorSetLayout) {
SizeT purgedSets = 0;
for (auto& frame : m_frames) {
@@ -1149,16 +1326,58 @@ namespace MobileGL::MG_Backend::DirectVulkan {
state.BoundShaderImages[unit].InternalFormat));
if (format == VK_FORMAT_UNDEFINED) format = MG_Util::ConvertTextureInternalFormatToVkEnum(internalFormat);
const auto feature = storage ? VK_FORMAT_FEATURE_STORAGE_TEXEL_BUFFER_BIT : VK_FORMAT_FEATURE_UNIFORM_TEXEL_BUFFER_BIT;
if (format == VK_FORMAT_UNDEFINED || !BufferFormatSupportsFeature(m_physicalDevice, format, feature))
WireDescriptorFatal("texel-buffer-native-format@P7");
if (format == VK_FORMAT_UNDEFINED || !BufferFormatSupportsFeature(m_physicalDevice, format, feature)) {
// P7 A.5, a decline (rule I (a)). GL's buffer-texture format list is wider than what
// Vulkan mandates for texel buffers - GL_RGB32{F,I,UI} and the 16-bit UNORM members
// are the realistic misses on a real driver - so this is a device capability gap, not
// anything the application did wrong, and nothing it could have sent differently.
//
// The placeholder is the RIGHT answer here, where it was not for the unaligned range
// above, and the difference is whether a correct answer exists at all. There, the
// bytes are addressable and only the descriptor cannot name them, so a placeholder
// would have hidden a fixable gap behind silently dropped writes. Here the device
// cannot represent this format as a texel buffer in any form - so the honest
// observable is monolith's for a missing view: the fetch reads zeros and the draw
// survives. AcquireUnboundTexelBufferView falls back to the R32 member of the
// shader's numeric class, whose texel-buffer support is mandatory, so the fallback
// itself cannot fail for want of device features.
MGLOG_E_ONCE("ResolveWireTexelBufferDescriptor: binding %u wants VkFormat %d as a %s texel buffer and "
"this device does not support it; the binding falls back to the zero placeholder and the "
"fetch reads zeros (texel-buffer-native-format)",
binding, static_cast<Int>(format), storage ? "storage" : "uniform");
return placeholder();
}
const VkDeviceSize texelSize = MG_Util::GetSizedInternalFormatSizeInBytes(internalFormat);
VkDeviceSize start = 0, size = 0;
if (texelSize == 0 || !ResolveWireRange(record.Desc.BufOffset, record.Desc.BufSize, slice.size, start, size)) return false;
size = std::min(size / texelSize, static_cast<VkDeviceSize>(m_wireMaxTexelElements)) * texelSize;
if (size == 0) return false;
if (start > std::numeric_limits<VkDeviceSize>::max() - slice.offset ||
(slice.offset + start) % m_wireTexelOffsetAlignment != 0)
WireDescriptorFatal("unaligned-texel-buffer-range@P7");
if (start > std::numeric_limits<VkDeviceSize>::max() - slice.offset)
WireDescriptorFatal("texel-buffer-offset-overflow");
if ((slice.offset + start) % m_wireTexelOffsetAlignment != 0) {
// P7 A.4, a named decline (rule I (a)). Not reachable through the public API:
// glTexBufferRange already refuses an offset that is not a multiple of
// GL_TEXTURE_BUFFER_OFFSET_ALIGNMENT (GL_Texture.cpp), and that cap is published
// from this same minTexelBufferOffsetAlignment. It stays a decline rather than a
// copy for the reason below.
const Bool writable = storage && state.BoundShaderImages[unit].Access != kMGPipeImageAccessReadOnly;
MGLOG_E_ONCE("ResolveWireTexelBufferDescriptor: binding %u is bound at offset %llu, which is not a "
"multiple of this device's minTexelBufferOffsetAlignment (%llu); the %s view is DECLINED "
"(unaligned-texel-buffer-range)", binding,
static_cast<unsigned long long>(slice.offset + start),
static_cast<unsigned long long>(m_wireTexelOffsetAlignment),
writable ? "writable" : "read-only");
// A read-only fetch can still be served the way an unbound buffer texture is -
// zeros, which is what a missing view gives in monolith - and that keeps the draw.
// A writable imageBuffer must not: the placeholder is shared by every unbound
// binding, so absorbing real stores into it would alias other bindings AND lose
// the application's writes without a word. Copying into an aligned transient slice
// is not the answer either: the transient arena carries no
// {UNIFORM,STORAGE}_TEXEL_BUFFER usage (VkBufferManager::InitializeTransientArenas),
// so no VkBufferView can be made over it, and widening a shared per-frame
// allocation to buy an unreachable path is the wrong trade. See notes/p7/magma-a.md.
return writable ? false : placeholder();
}
VkBufferViewCreateInfo viewInfo{VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO};
viewInfo.buffer = slice.buffer;
viewInfo.format = format;
@@ -1546,11 +1765,56 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (!m_bufferManager->AcquireWireSlice(BufferKind::ShaderStorage, range.Res, slice) || !slice.IsValid()) return false;
VkDeviceSize start = 0, size = 0;
if (!ResolveWireRange(range.Offset, range.Size, slice.size, start, size)) return false;
if (start > std::numeric_limits<VkDeviceSize>::max() - slice.offset ||
(slice.offset + start) % m_wireStorageOffsetAlignment != 0)
WireDescriptorFatal(isAtomicCounterBlock ? "unaligned-atomic-buffer-range@P7"
: "unaligned-storage-buffer-range@P7");
if (size > m_wireMaxStorageRange) WireDescriptorFatal("storage-buffer-native-range@P7");
if (start > std::numeric_limits<VkDeviceSize>::max() - slice.offset)
WireDescriptorFatal("storage-buffer-offset-overflow");
if ((slice.offset + start) % m_wireStorageOffsetAlignment != 0) {
// P7 A.4, a named decline (rule I (a)), and the one shape in this cluster an
// application can actually reach. GL has no queryable atomic-counter buffer
// offset alignment, so glBindBufferRange only enforces offset % 4 for
// GL_ATOMIC_COUNTER_BUFFER (GL 4.6 core 6.1.1, GL_Buffer.cpp
// ValidateBufferRangeOffsetAndSize); glslang lowers the counter block onto a
// storage buffer, whose descriptor offset must be a multiple of
// minStorageBufferOffsetAlignment - 16 on lavapipe, 64 on Adreno. Offset 4 is
// therefore a legal bind this backend cannot express. The plain SSBO half is
// unreachable by comparison: GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT is
// published from that same Vulkan limit, so the frontend already refused it
// with GL_INVALID_VALUE.
//
// DECLINED RATHER THAN COPIED, deliberately. Copying the window into an aligned
// transient slice reads correctly and then silently DROPS every shader write,
// and this site cannot tell a writable block from a read-only one: there is no
// readonly bit in the storage-block reflection (ProgramFactory.h carries only
// name and index), an atomic counter is written by definition, and
// MarkWireBufferGpuWritten below already assumes any binding here may be
// written. A correct copy needs a post-dispatch copy-back from the transient
// slice to the unaligned source, ordered with a barrier - and the place to
// record that is the draw/dispatch tail in VulkanRenderer.cpp / WireDraw.inc,
// which belong to wave-2 packages B and C, not to this file. Losing the
// dispatch with a named line in the role log beats returning stale counters
// with no error at all. notes/p7/magma-a.md carries the recipe.
MGLOG_E_ONCE("ResolveStorageBufferDescriptor: block '%s' is bound at offset %llu, which is not a "
"multiple of this device's minStorageBufferOffsetAlignment (%llu); %s ranges cannot be "
"expressed as a descriptor and the shader's writes would be dropped by a read-only "
"copy, so the binding is DECLINED (unaligned-%s-buffer-range)",
blockName.c_str(), static_cast<unsigned long long>(slice.offset + start),
static_cast<unsigned long long>(m_wireStorageOffsetAlignment),
isAtomicCounterBlock ? "atomic-counter" : "shader-storage",
isAtomicCounterBlock ? "atomic" : "storage");
return false;
}
if (size > m_wireMaxStorageRange) {
// P7 A.3: clamp, do not refuse. The monolith arm binds whatever the GL range
// asked for and never consults maxStorageBufferRange at all (see the resident
// path below), so refusing here was the wire arm inventing a death the other
// backend does not have. A clamped binding keeps every byte the device can
// name reachable; past it the shader is out of the descriptor and
// robustBufferAccess answers zero, which beats losing the draw outright.
MGLOG_E_ONCE("ResolveStorageBufferDescriptor: block '%s' bound %llu bytes, past this device's "
"maxStorageBufferRange of %llu; the binding is clamped to the limit",
blockName.c_str(), static_cast<unsigned long long>(size),
static_cast<unsigned long long>(m_wireMaxStorageRange));
size = m_wireMaxStorageRange;
}
outBufferInfo = {slice.buffer, slice.offset + start, size};
// Shader writes stay in the canonical server VkBuffer. The event marks the
// client's shadow stale; neither this path nor later acquires seed it back.
@@ -2358,9 +2622,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint32 blockIndex, Uint32 bindingPoint, UboBindResult& out) const {
if (bindingPoint >= MG_Pipe::kMGPipeMaxBufferBindingPoints)
WireDescriptorFatal("uniform-buffer-binding-point");
const VkDeviceSize blockSize = program.GetUBOSizeAt(blockIndex);
VkDeviceSize blockSize = program.GetUBOSizeAt(blockIndex);
if (blockSize == 0) return false;
if (blockSize > m_wireMaxUniformRange) WireDescriptorFatal("uniform-block-native-range@P7");
if (blockSize > m_wireMaxUniformRange) {
// P7 A.3: clamp, do not refuse. A conformant program cannot get here -
// GL_MAX_UNIFORM_BLOCK_SIZE is published straight from this same
// maxUniformBufferRange (BackendLoaders/Vulkan/Loader.cpp), so a block this big
// fails to link long before a draw - which makes refusing it a Fatal for a shape
// nothing can produce. If the two ever disagree, the honest answer is the bytes
// the device CAN name: everything past the clamp is outside the descriptor and
// reads back as zero under robustBufferAccess, which is inside GL's "undefined"
// for a block the implementation never promised to hold. Clamping keeps the range
// constant across draws, so the descriptor-set reuse hash is unaffected.
MGLOG_E_ONCE("ResolveWireUniformBufferPayload: reflected uniform block %u is %llu bytes, past this "
"device's maxUniformBufferRange of %llu; the binding is clamped to the limit and the "
"remainder reads zero",
blockIndex, static_cast<unsigned long long>(blockSize),
static_cast<unsigned long long>(m_wireMaxUniformRange));
blockSize = m_wireMaxUniformRange;
}
const auto& range = MG_Pipe::MGPipeApplier().BoundShaderBuffers[
MG_Pipe::kMGPipeShaderBufferClassUniform][bindingPoint];
BufferSlice source{};
@@ -2387,17 +2667,45 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// inside the GL binding range. A readback here could retire the command buffer
// whose value BindProgramUniformBuffers is currently holding.
const VkDeviceSize copied = std::min(available, blockSize);
if (bound && ((start & 3u) != 0 || ((source.offset + start) & 3u) != 0 || (copied & 3u) != 0))
WireDescriptorFatal("uniform-buffer-byte-tail@P7");
static thread_local Vector<Uint8> zero;
zero.assign(static_cast<SizeT>(blockSize), 0);
BufferSlice padded{};
if (!m_bufferManager->UploadTransient(BufferKind::Uniform, m_wireFrameIndex, zero.data(), blockSize,
std::max<VkDeviceSize>(4, m_minDynamicOffsetAlignment), padded) || !padded.IsValid()) return false;
if (bound && copied != 0 &&
!m_bufferManager->CopyWireBufferRangeToSlice(range.Res, start, copied, padded)) return false;
if (padded.offset > std::numeric_limits<Uint32>::max())
WireDescriptorFatal("uniform-buffer-dynamic-offset@P7");
if (bound && copied != 0) {
// P7 A.1 retires `uniform-buffer-byte-tail`. The window is word-aligned at both
// ends on nearly every bind - glBindBufferRange already forces the offset onto
// GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, and that is a multiple of four on every
// device - but GL 4.6 core 6.1.1 puts no such rule on the SIZE, so `copied` is free
// to end mid-word and the Redmi Minecraft run lands there
// (notes/p5f/magma-inproc-fix.md #5). An unaligned start is not reachable through
// the public API and only a protocol-level offset can produce one; the sub-word arm
// handles it anyway rather than leaving a second shape to trip over later. Both arms
// copy the same bytes to the same place: the sub-word one pays one extra transient
// slice and one extra region copy, and nothing outside [start, start + copied)
// reaches the block, which is what makes the padded zeros the visible tail.
const Bool wordWindow = ((start | copied) & 3u) == 0;
if (!(wordWindow
? m_bufferManager->CopyWireBufferRangeToSlice(range.Res, start, copied, padded)
: m_bufferManager->CopyWireBufferSubWordRangeToSlice(range.Res, start, copied,
m_wireFrameIndex, padded, 0)))
return false;
}
if (padded.offset > std::numeric_limits<Uint32>::max()) {
// P7 A.2, a decline (rule I (a)), not a session death. A Vulkan dynamic offset is a
// uint32_t, so a transient uniform ring that grew past 4 GiB inside one frame has
// nowhere to put this block - which is our ring's problem, not a protocol fault, and
// nothing the peer could have sent differently. One binding is skipped and the draw
// is lost; the session keeps running and the next frame's rewound ring resolves it.
// The monolith arm has the same ceiling and truncates silently on the way out
// (ResolveDynamicUboDescriptor's static_cast<Uint32>), i.e. it binds a WRONG window
// rather than none, so declining is the strictly better observable of the two.
MGLOG_E_ONCE("ResolveWireUniformBufferPayload: the transient uniform ring reached offset "
"%llu, past the 4 GiB a Vulkan dynamic offset can name; uniform binding "
"point %u is declined for this draw",
static_cast<unsigned long long>(padded.offset), bindingPoint);
return false;
}
out.directBindable = true;
out.buffer = padded.buffer;
out.range = blockSize;
@@ -2476,9 +2784,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
auto& bindingPoint = MGB_CTX->GetBufferBindingPoint(BufferTarget::Uniform, frontendBinding);
const auto& bufferObject = bindingPoint.GetBoundObject();
MOBILEGL_ASSERT(bufferObject != nullptr,
"ResolveUniformBufferPayload: no UBO bound at frontend binding %u for block '%s'",
frontendBinding, program.GetUniformBlockName(static_cast<Uint32>(blockIndex)).c_str());
static thread_local Vector<Uint8> paddedUbo;
if (bufferObject == nullptr) {
// A block the program declares at a point with nothing bound - never bound, or its
// buffer was deleted while bound, which unbinds it (UnboundUniformBlockScenario,
// DeletedBoundBufferScenario). GL leaves what the shader READS undefined and raises no
// error; this used to dereference the null object. Answer it with a zeroed block of the
// reflected size, as the wire arm does (ResolveWireUniformBufferPayload).
const VkDeviceSize blockSize =
static_cast<VkDeviceSize>(program.GetUBOSizeAt(static_cast<Uint32>(blockIndex)));
paddedUbo.assign(static_cast<SizeT>(blockSize > 0 ? blockSize : 16), 0);
out.payload = paddedUbo.data();
out.payloadSize = static_cast<VkDeviceSize>(paddedUbo.size());
return true;
}
bufferObject->SyncPersistentMappedRange();
MOBILEGL_ASSERT(bufferObject->MappedData() != nullptr && bufferObject->GetSize() != 0,
@@ -2511,7 +2830,6 @@ namespace MobileGL::MG_Backend::DirectVulkan {
outSize = blockSize;
outData = bufferObject->MappedData() + static_cast<SizeT>(rangeStart);
if (available < blockSize) {
static thread_local Vector<Uint8> paddedUbo;
paddedUbo.assign(static_cast<SizeT>(blockSize), 0);
Memcpy(paddedUbo.data(), outData, static_cast<SizeT>(available));
outData = paddedUbo.data();
@@ -2607,32 +2925,49 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return true;
}
Uint32 UniformManager::NextDescriptorPoolMaxSets(const Vector<DescriptorPoolBucket>& pools, Bool updateAfterBind,
Uint32 initialMaxSets) {
const Uint32 minSets = std::max<Uint32>(1, initialMaxSets);
const Uint32 maxSets = std::max(minSets, kMaxDescriptorPoolSets);
Uint32 largest = 0;
for (const auto& pool : pools) {
if (pool.updateAfterBind == updateAfterBind) {
largest = std::max(largest, pool.maxSets);
}
}
if (largest == 0) {
return minSets;
}
const Uint64 doubled = static_cast<Uint64>(largest) * 2;
return static_cast<Uint32>(std::clamp<Uint64>(doubled, minSets, maxSets));
}
Bool UniformManager::GrowFrameDescriptorPool(FrameResources& frame, Uint32 frameIndex, Bool updateAfterBind) {
if (frame.descriptorPools.empty()) {
return false;
}
const auto matchingBucket = std::find_if(
frame.descriptorPools.begin(), frame.descriptorPools.end(),
[updateAfterBind](const DescriptorPoolBucket& candidate) { return candidate.updateAfterBind == updateAfterBind; });
const Uint32 currentMaxSets = matchingBucket != frame.descriptorPools.end()
? std::max<Uint32>(1, matchingBucket->maxSets)
: m_setsPerFrame;
const Uint32 grownMaxSets = currentMaxSets <= (std::numeric_limits<Uint32>::max() / 2) ? (currentMaxSets * 2)
: currentMaxSets;
Uint32 grownMaxSets = NextDescriptorPoolMaxSets(frame.descriptorPools, updateAfterBind, m_setsPerFrame);
VkDescriptorPool grownPool = VK_NULL_HANDLE;
if (!CreateDescriptorPool(grownMaxSets, updateAfterBind, grownPool)) {
MGLOG_E_ONCE("UniformDescriptorBinder::GrowFrameDescriptorPool failed: cannot create grown pool (%u -> %u sets)",
currentMaxSets, grownMaxSets);
return false;
// A large pool is a large driver allocation; if the device refuses one, settle for less
// rather than failing the draw - halving down to the base size, where the old policy lived.
while (!CreateDescriptorPool(grownMaxSets, updateAfterBind, grownPool)) {
if (grownMaxSets <= m_setsPerFrame) {
MGLOG_E_ONCE("UniformDescriptorBinder::GrowFrameDescriptorPool failed: cannot create grown pool (%u sets)",
grownMaxSets);
return false;
}
grownMaxSets = std::max(m_setsPerFrame, grownMaxSets / 2);
}
frame.descriptorPools.push_back({grownPool, grownMaxSets, 0, updateAfterBind});
frame.activeDescriptorPoolIndex = static_cast<Uint32>(frame.descriptorPools.size() - 1);
MGLOG_D(
"UniformDescriptorBinder: frame %u descriptor pool exhausted, grew pool (%u -> %u sets), poolCount=%zu",
frameIndex, currentMaxSets, grownMaxSets, frame.descriptorPools.size());
frame.capacitySets += grownMaxSets;
// Growing is the opposite of quiet: a trim must wait a full quiet run from here.
frame.quietEpochs = 0;
MGLOG_D("UniformDescriptorBinder: frame %u descriptor pool exhausted, grew a %u-set pool (updateAfterBind=%d), "
"poolCount=%zu",
frameIndex, grownMaxSets, updateAfterBind ? 1 : 0, frame.descriptorPools.size());
return true;
}
@@ -2809,7 +3144,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool UniformManager::PrepareWireTextureResources(const MagmaProgramSource& program,
const ProgramFactory::VkProgramObject& programObj) {
if (!program.IsWire()) return true;
if (programObj.declinedDescriptors || m_textureManager == nullptr) return false;
if (programObj.declinedDescriptors || m_textureManager == nullptr) {
MGL_WIRE_DECLINE_AT(DescriptorDeclined, "the program carries a descriptor MobileGL could not resolve");
return false;
}
const auto& state = MG_Pipe::MGPipeApplier();
// Resolve writable images first: aliases sampled by the same draw must
// see the final STORAGE-capable allocation before any view is built.
@@ -2823,16 +3161,28 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const Uint32 count = BindingDescriptorCount(programObj, binding);
for (Uint32 element = 0; element < count; ++element) {
const Int location = baseLocation + static_cast<Int>(element);
if (baseLocation < 0 || !program.UniformLocationsAliasSameUniform(baseLocation, location))
if (baseLocation < 0 || !program.UniformLocationsAliasSameUniform(baseLocation, location)) {
MGL_WIRE_DECLINE_AT(SamplerLocationAlias,
"binding %u element %u does not alias one sampler uniform", binding,
element);
return false;
}
const Int unit = program.GetUniformSamplerOrImageUnitIndex(static_cast<Uint>(location));
if (unit < 0 || static_cast<Uint32>(unit) >=
(storage ? MG_Pipe::kMGPipeMaxImageUnits : MG_Pipe::kMGPipeMaxTextureUnits)) return false;
(storage ? MG_Pipe::kMGPipeMaxImageUnits : MG_Pipe::kMGPipeMaxTextureUnits)) {
MGL_WIRE_DECLINE_AT(SamplerUnitRange, "sampler/image unit %d is out of range", unit);
return false;
}
const auto handle = storage ? state.BoundShaderImages[unit].Res
: state.BoundSamplerViews[unit].Texture;
// Null is the wire's unbound/incomplete binding. Its native
// placeholder needs no resource upload or frontend allocation.
if (MG_Pipe::MGPipeHandleIsNull(handle)) continue;
// Nor does an image unit naming a (level, layer) its texture lacks: the
// descriptor resolve binds that same placeholder (GL 4.6 core 8.26), and
// preparing the texture would decline a storage-less one and lose the pass.
if (storage && WireShaderImageNamesNoTexel(*m_textureManager, state.BoundShaderImages[unit]))
continue;
if (!m_textureManager->SyncTextureResourceByHandle(handle, false, storage)) return false;
}
}
@@ -2880,6 +3230,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
(samplerBindingOverrides == nullptr || samplerBindingOverrides->empty());
if (cacheable && samplerDescriptorsUnchangedHint && m_fastRebindMemo.valid &&
m_fastRebindMemo.frameIndex == frameIndex &&
#if MOBILEGL_BUILD_DISAGGREGATED
// A wire store destroyed since the memo was taken may have handed its handle
// value to a later mint (see FastRebindMemo): "same VkBuffer" then names a
// different store, so the memo is refused and the full walk re-records it.
m_fastRebindMemo.wireStoreDestroyEpoch == m_bufferManager->GetWireStoreDestroyEpoch() &&
#endif
m_fastRebindMemo.programLifetimeId == program.GetLifetimeId() &&
m_fastRebindMemo.programHash == programObj.hash) {
VkBuffer uboBuffer = VK_NULL_HANDLE;
@@ -3176,6 +3532,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
};
mixWords(&programObj.descriptorSetLayout, sizeof(programObj.descriptorSetLayout));
#if MOBILEGL_BUILD_DISAGGREGATED
// P7 M2 round 2 (ID-P7-43): the buffer infos below name WIRE stores by VkBuffer
// handle, and a wire store can be destroyed mid-frame (VkBufferManager::
// DeferredWireRelease) with its handle value re-minted before this frame's memo
// is cleared. Folding the destroy epoch in makes every entry taken before such a
// destroy miss, so a byte-identical info can never revive a set baked to a dead
// store. NOT the slice epoch: that moves on every glBufferSubData and would make
// the memo miss on every draw.
mix64(m_bufferManager->GetWireStoreDestroyEpoch());
#endif
for (const auto& write : writes) {
mix64((static_cast<Uint64>(write.dstBinding) << 40) ^
(static_cast<Uint64>(write.descriptorType) << 8) ^
@@ -3230,7 +3596,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_fastRebindMemo = FastRebindMemo{
/*valid=*/true, frameIndex, program.GetLifetimeId(), programObj.hash,
fastRebindUboBinding, bufferInfos[0].buffer,
bufferInfos[0].range, descriptorSet};
bufferInfos[0].range, descriptorSet,
#if MOBILEGL_BUILD_DISAGGREGATED
m_bufferManager->GetWireStoreDestroyEpoch(),
#endif
};
} else {
m_fastRebindMemo.valid = false;
}
@@ -13,6 +13,8 @@
#include "VkBufferManager.h"
#include "VkSamplerManager.h"
#include "VkTextureManager.h"
#include "WirePlaceholderKey.h"
#include "../DescriptorPoolCensus.h"
#include "../VkIncludes.h"
#include <Includes.h>
@@ -54,6 +56,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void Shutdown();
void BeginFrame(Uint32 frameIndex);
#if MOBILEGL_BUILD_DISAGGREGATED
// A present-less wire frame may run arbitrarily many draws. Before its
// descriptor cursor reaches this budget, the renderer retires the
// submission that last used the sets and rewinds every layout cursor.
static constexpr Uint32 kWireDescriptorSetBudget = 2048;
Bool WireDescriptorSetBudgetReached(Uint32 frameIndex) const;
SizeT RewindWireDescriptorSets(Uint32 frameIndex);
// How the wire arm answers a STORAGE image unit that GL 4.6 core 8.26 makes invalid (no
// texture, or a (level, layer) its texture lacks): a load returns zero, a store and an
// atomic are discarded. `deviceNullDescriptor` is the renderer's enablement of
// VK_EXT_robustness2's nullDescriptor at device creation. With it the unit binds a NULL
// storage-image descriptor, which is exactly that rule. Without it (or under
// MGITEST_MAGMA_FORCE_PRIVATE_IMAGE_PLACEHOLDER=1, the host lanes' way onto this arm) the
// unit binds a placeholder PRIVATE to its unit (per shape), cleared before each use, so no
// two units ever alias; what that cannot give is a load of the SAME unit after its own
// store inside one pass reading zero, nor privacy past kWirePrivateStoragePlaceholderCap
// placeholders (CONTRACT-P7 §12). Called after Initialize.
void SetWireInvalidStorageImageArm(Bool deviceNullDescriptor);
#endif
// A command buffer (re)began recording: descriptor bindings recorded into
// the previous buffer do not carry over, so drop the bind-dedup shadow.
void OnCommandBufferBoundary() { m_lastBindValid = false; }
@@ -64,9 +85,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// FREE_DESCRIPTOR_SET_BIT) and the pool accounting is credited, so program
// churn recycles pool capacity instead of abandoning it. GPU-safe: the layout
// only dies after >1024 idle frame boundaries, so no in-flight command buffer
// references its sets. This is the only eviction path for the per-layout
// caches - a live layout's entry must never be purged (its sets would be
// unreachable pool slots), so there is deliberately no age-based sweep here.
// references its sets. This is the only PER-LAYOUT eviction path for the
// caches - a live layout's entry must never be purged on its own (its sets would be
// unreachable pool slots), so there is deliberately no age-based sweep here. The
// other path, TrimFrameDescriptorPools, drops a slot's whole cache together with
// the pools behind it.
void OnDescriptorSetLayoutDestroyed(VkDescriptorSetLayout descriptorSetLayout);
// One record per visited CombinedImageSampler DESCRIPTOR (post fallback substitution,
// in binding order, and within a binding in array-element order): the resolved texture
@@ -138,13 +161,46 @@ namespace MobileGL::MG_Backend::DirectVulkan {
static Bool ProgramSamplesOnlySingleLevelTextures(const MagmaProgramSource& program,
const ProgramFactory::VkProgramObject& programObj);
private:
// ---- per-frame descriptor pools ----------------------------------------------------
// Each frame slot owns a list of pools. The first is the base pool (non-update-after-bind,
// setsPerFrame sets) and is never destroyed while the manager lives; every later one was
// added by GrowFrameDescriptorPool when an allocation found no room. A pool serves one
// updateAfterBind flavour, because a set can only come from a pool created with the flag
// its layout was.
struct DescriptorPoolBucket {
VkDescriptorPool handle = VK_NULL_HANDLE;
Uint32 maxSets = 0;
Uint32 allocatedSets = 0;
Bool updateAfterBind = false;
};
// One pool's maxSets never exceeds this. It bounds one pool's driver memory: CreateDescriptorPool
// declares 8 descriptors of each of 6 types per set, so a 4096-set pool asks for 196,608
// descriptors - about 12 MiB at 64 bytes each on a driver that backs every declared one.
static constexpr Uint32 kMaxDescriptorPoolSets = 4096;
// The growth policy, pure so the host tests can drive it: the size of the next pool a slot
// adds for `updateAfterBind` layouts, given the slot's existing pools. Twice the LARGEST pool
// of that flavour, capped at kMaxDescriptorPoolSets; `initialMaxSets` when the slot has no
// pool of that flavour yet. So a frame that needs N sets adds O(log N) pools while N stays
// under the cap, and one more pool per kMaxDescriptorPoolSets sets beyond it. Taking the
// largest (not the first) pool is the point: the first same-flavour pool is the oldest,
// smallest one, and doubling IT made every grown pool the same size.
static Uint32 NextDescriptorPoolMaxSets(const Vector<DescriptorPoolBucket>& pools, Bool updateAfterBind,
Uint32 initialMaxSets);
// Trim policy (see TrimFrameDescriptorPools). A slot is "quiet" in an epoch whose peak set
// count times kDescriptorTrimHeadroom still fits its capacity; a slot that has been quiet for
// MOBILEGL_MAGMA_DESCRIPTOR_TRIM_FRAMES consecutive epochs, and holds at least
// kDescriptorTrimMinCapacity sets, is trimmed. The headroom is what keeps the trim from
// thrashing: geometric growth leaves a re-grown slot at least a third full (the pool a growth
// adds is at most twice the largest before it, so at most twice everything the slot held, and
// the epoch that grew it needed more than that), so the epoch that grew a slot is never quiet
// for it. The floor keeps a slot sized for ordinary
// frames from being trimmed at all - the trim exists to give back a burst.
static constexpr Uint32 kDescriptorTrimHeadroom = 4;
static constexpr Uint32 kDescriptorTrimMinCapacity = 1024;
// For tests (DescriptorPoolCensus.h): pools, capacity, cached sets and trims over every slot.
DescriptorPoolCensus GetDescriptorPoolCensus() const;
private:
// A cached descriptor set together with the pool it was allocated from, so a
// layout-destroyed purge can vkFreeDescriptorSets it back and credit the
@@ -202,13 +258,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkImageLayout layout = VK_IMAGE_LAYOUT_UNDEFINED;
Uint32 layers = 1;
};
mutable UnorderedMap<Uint64, WirePlaceholderImage> m_wirePlaceholderImages;
// Keyed by shape, and a storage placeholder on the arm without a null descriptor by its
// image unit too (WirePlaceholderKey.h: ChooseWirePlaceholderKey, and its cap).
mutable UnorderedMap<WirePlaceholderKey, WirePlaceholderImage, WirePlaceholderKeyHash> m_wirePlaceholderImages;
// How many of those are private to a unit (against kWirePrivateStoragePlaceholderCap).
mutable SizeT m_wirePrivateStoragePlaceholders = 0;
// See SetWireInvalidStorageImageArm.
Bool m_wireInvalidStorageImagesBindNull = false;
// boundStorageFormat: for a storage binding with no reflected format, the format the
// unit's glBindImageTexture named, when the unit holds a texture it cannot address
// (GL 4.6 core 8.26); UNDEFINED keeps the numeric-domain R32 default. `unit` is the image
// unit of a storage binding (what its private placeholder is private to); unused for a sampler.
Bool ResolveWirePlaceholderImage(VkCommandBuffer commandBuffer, const MagmaProgramSource& program,
const ProgramFactory::VkProgramObject& programObj, Uint32 binding, Bool storage,
VkDescriptorImageInfo& out) const;
VkDescriptorImageInfo& out, VkFormat boundStorageFormat = VK_FORMAT_UNDEFINED, Uint32 unit = ~0u) const;
void DestroyWirePlaceholderImage(WirePlaceholderImage& image) const;
#endif
// descriptorSetCacheByLayout is the per-frame set cache: BeginFrame rewinds every cursor, so
// a slot re-writes the sets it allocated in earlier frames instead of allocating new ones, and
// the cache only ever grows - except through TrimFrameDescriptorPools, which gives a slot's
// whole burst back at once (and OnDescriptorSetLayoutDestroyed, one dead layout at a time).
struct FrameResources {
Vector<DescriptorPoolBucket> descriptorPools;
UnorderedMap<VkDescriptorSetLayout, DescriptorSetCacheEntry> descriptorSetCacheByLayout;
@@ -222,6 +292,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint32 activeDescriptorPoolIndex = 0;
Uint32 allocatedSetsThisFrame = 0;
Uint32 peakAllocatedSetsThisFrame = 0;
// Sum of maxSets over descriptorPools, kept in step by Initialize, Grow and Trim.
Uint64 capacitySets = 0;
// Consecutive epochs (see NoteEndedEpochForTrim) that were quiet for THIS slot's capacity.
Uint32 quietEpochs = 0;
};
static Bool ResolveSamplerTexture(const MagmaProgramSource& program,
@@ -365,6 +439,28 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const Vector<Uint32>& dynamicOffsets);
Bool CreateDescriptorPool(Uint32 maxSets, Bool updateAfterBind, VkDescriptorPool& outPool) const;
Bool GrowFrameDescriptorPool(FrameResources& frame, Uint32 frameIndex, Bool updateAfterBind);
// An epoch is the stretch between two BeginFrame calls: one frame, or the part of one between
// two drains (VulkanRenderer::TryDrainFrameTransients). Called by BeginFrame with the epoch
// that just ended, it updates every slot's quiet count - a burst in any slot counts against
// all of them, so bursts that land in the other slots still hold a slot's trim off.
void NoteEndedEpochForTrim(Uint32 epochPeakSets);
// Gives a slot's whole descriptor burst back: clears its per-layout cache, destroys every
// grown pool and resets the base pool, so the slot re-grows geometrically from setsPerFrame.
// Whole-slot rather than per-set: vkFreeDescriptorSets fragments a pool and can never empty
// a grown one. Called only from BeginFrame, which runs only when no pending command buffer
// references the slot's sets:
// - Present path (VulkanRenderer::Present, after WaitAndAcquireNextImage): the slot's
// imageInFlightFence has been waited. That fence rides the slot's LAST submission - the
// Present submit - and every mid-frame submission of the slot (FlushPendingCommands,
// SplitOversizedRecording, readbacks) went to the same queue earlier, so the fence's
// signal covers them (vkQueueSubmit's fence scope includes all earlier submissions).
// - Drain path (VulkanRenderer::TryDrainFrameTransients): every submission has been
// observed complete and no recording is open.
// Sets are per slot: a slot's sets are only ever bound into recordings made while it is the
// current slot, so trimming one slot never touches another slot's in-flight work.
void TrimFrameDescriptorPools(FrameResources& frame, Uint32 frameIndex);
// Drops every handle this class holds to a descriptor set outside the per-layout caches.
void InvalidateDescriptorSetMemos();
VkResult AllocateDescriptorSetsFromActivePool(
Uint32 frameIndex, const ProgramFactory::VkProgramObject& programObj, VkDescriptorSet& outDescriptorSet);
VkResult AcquireDescriptorSet(Uint32 frameIndex,
@@ -382,6 +478,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint32 m_maxBindings = 0;
Uint32 m_setsPerFrame = 0;
Uint32 m_peakDescriptorSetsObserved = 0;
// MOBILEGL_MAGMA_DESCRIPTOR_TRIM_FRAMES, latched at Initialize (0 = never trim).
Uint32 m_trimQuietEpochs = 0;
// The slot whose epoch is open (the last one BeginFrame was called for); valid once
// m_epochOpen is set. Its peak is the ended epoch's peak at the next BeginFrame.
Uint32 m_epochSlot = 0;
Bool m_epochOpen = false;
Uint64 m_descriptorSlotTrims = 0;
VkTextureManager* m_textureManager = nullptr;
VkSamplerManager* m_samplerManager = nullptr;
mutable SharedPtr<MG_State::GLState::ITextureObject> m_fallbackTexture2D;
@@ -442,6 +545,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// drops the hint upstream; an arena wrap or growth resolves a different
// VkBuffer and misses. AcquireDescriptorSet's per-frame cursor only
// advances, so the recorded set is never re-written within its frame.
// P7 M2 round 2 (ID-P7-43): a WIRE store can die MID-FRAME on the reclaim
// path (VkBufferManager::DeferredWireRelease), and the next mint can hand
// its VkBuffer handle value back - a heap pointer under lavapipe - so "the
// same VkBuffer+range" is no longer proof of the same store. The memo
// therefore records VkBufferManager::GetWireStoreDestroyEpoch() and misses
// once any wire store has been destroyed since; the descriptor-reuse
// signature below folds the same epoch in. On the wire arm this memo is
// never consulted (SetupWireDraw passes no sampler hint), so the field is
// the defensive half; the signature is the half the wire arm reaches.
struct FastRebindMemo {
Bool valid = false;
Uint32 frameIndex = 0;
@@ -451,6 +563,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkBuffer uboBuffer = VK_NULL_HANDLE;
VkDeviceSize uboRange = 0;
VkDescriptorSet set = VK_NULL_HANDLE;
#if MOBILEGL_BUILD_DISAGGREGATED
Uint64 wireStoreDestroyEpoch = 0;
#endif
};
FastRebindMemo m_fastRebindMemo;
@@ -442,41 +442,94 @@ namespace MobileGL::MG_Backend::DirectVulkan {
for (Uint32 location = 0; location < elements.AttributeCount; ++location) {
const auto& attr = elements.Attributes[location];
if (!attr.Enabled || !(activeMask & (1u << location))) continue;
const auto reject = [&](const char* reason, VkFormat format = VK_FORMAT_UNDEFINED) {
MGLOG_E("Magma wire vertex layout: %s location=%u type=%u size=%u normalized=%u integer=%u long=%u bgra=%u stride=%d offset=%llu format=%d bufferWindow=%u+%u activeMask=0x%x",
// P7 wave 2 package C, CONTRACT-P7 §3.2 `vertex-layout`: THE ONE STRING BECAME TWO
// VERDICTS, and the split is by what the reason is ABOUT rather than by severity.
//
// Every reason here used to return false, and the single caller answered that with
// one P7-marked MagmaWireFatal - an abort, bypassing Session::Fail, invisible
// to the census gate and with no equivalent on the monolith arm at all. But the
// reasons are not one kind of thing:
//
// * `buffer-window` is a statement about the RECORD: an enabled attribute the
// program reads names a slot outside the window set_vertex_buffers published.
// Nothing about the device or the format is involved; the two halves of the
// protocol disagree about what crossed. It stays a named Fatal through the hook
// (see the caller), and it is load-bearing rather than defensive - it is what
// caught the Redmi VAO hash collision fixed in 376c04be.
//
// * every other reason is a statement about what VULKAN CAN EXPRESS on this
// device: a GL type with no VkFormat, a format without
// VK_FORMAT_FEATURE_VERTEX_BUFFER_BIT, a component size we cannot compute, an
// offset sum that does not fit. The monolith arm has had an answer for these
// since it was written, and it is not an abort: mask the attribute out of the
// vertex input state and carry on (the `unsupportedAttribMask |= ...; continue;`
// sites in GetOrCreateVertexInputState above). The draw path then decides,
// loudly and once, whether the masked attribute was one the program actually
// reads. This arm now does exactly that, which is what "same observable as the
// monolith lane" means for this row.
//
// THE COUNT: the contract says "the other six"; there are SEVEN return sites below
// it, because `offset-overflow` was added after the audit's count and is an
// arithmetic guard rather than a shape one. It is masked with the rest: an
// attribute whose base offsets cannot be added is one Vulkan cannot fetch, the
// monolith arm never computes that sum at all, and masking is strictly safer than
// a Fatal for a number no application can reach on purpose. Recorded in
// notes/p7/magma-c.md rather than silently reconciled.
const auto describe = [&](const char* reason, VkFormat format) {
MGLOG_E_ONCE("Magma wire vertex layout: %s location=%u type=%u size=%u normalized=%u integer=%u long=%u bgra=%u stride=%d offset=%llu format=%d bufferWindow=%u+%u activeMask=0x%x",
reason, location, attr.Type, static_cast<Uint32>(attr.Size), static_cast<Uint32>(attr.Normalized),
static_cast<Uint32>(attr.IsInteger), static_cast<Uint32>(attr.IsLong), static_cast<Uint32>(attr.IsBgra),
attr.Stride, static_cast<unsigned long long>(attr.Offset), static_cast<Int>(format),
state.VertexBufferStart, state.VertexBufferCount, activeMask);
};
// The protocol verdict: the caller turns a false into the named Fatal.
const auto reject = [&](const char* reason, VkFormat format = VK_FORMAT_UNDEFINED) {
describe(reason, format);
return false;
};
// The device verdict: MGLOG_E_ONCE + mask + continue, the monolith arm's shape.
// MGLOG_E_ONCE and not MGLOG_E, also the monolith arm's: an unmappable attribute is
// a property of the VAO and the device, so it repeats every draw, and the per-draw
// line is what made the old abort look preferable to whoever wrote it.
const auto maskOut = [&](const char* reason, VkFormat format = VK_FORMAT_UNDEFINED) {
describe(reason, format);
out.unsupportedAttribMask |= 1u << location;
};
// set_vertex_buffers is flattened PER ATTRIBUTE (VertexInputEmit.h), not
// indexed by the original ARB binding point in attr.BindingIndex.
if (location < state.VertexBufferStart ||
location - state.VertexBufferStart >= state.VertexBufferCount) return reject("buffer-window");
const auto& buffer = state.VertexBuffers[location];
const auto type = static_cast<DataType>(attr.Type);
if (attr.Stride < 0 || attr.Size < 1 || attr.Size > 4) return reject("attribute-shape");
if (attr.Stride < 0 || attr.Size < 1 || attr.Size > 4) { maskOut("attribute-shape"); continue; }
VkFormat format = ToVkVertexFormat(type, attr.Size, attr.Normalized, attr.IsInteger,
attr.IsBgra, attr.IsLong);
auto conversion = VertexStreamConversion::None;
if (format == VK_FORMAT_UNDEFINED && type == DataType::Float64) {
const auto* backend = MG_Remote::Server::ServerLoopInstance().Backend();
if (backend && backend->GetDynamicParameters().SupportsFloat64VertexAttributes) return reject("native-fp64-format");
if (backend && backend->GetDynamicParameters().SupportsFloat64VertexAttributes) {
// Exactly the monolith arm's answer for the same state: with native fp64
// the module KEPT its 64-bit inputs (DemoteFloat64Pass did not run), so
// narrowing the stream would feed float32 to a Float64 input - and the
// monolith build therefore leaves sourceVkFormat UNDEFINED and falls into
// its own mask-out below. Same place, same mask.
maskOut("native-fp64-format");
continue;
}
format = ToFloat32VertexFormat(attr.Size);
conversion = VertexStreamConversion::Float64ToFloat32;
}
if (format == VK_FORMAT_UNDEFINED) return reject("format-map");
if (format == VK_FORMAT_UNDEFINED) { maskOut("format-map"); continue; }
if (!SupportsVertexBufferFormat(format)) {
if (!IsScaledIntegerVertexFormat(format)) return reject("native-format-feature", format);
if (!IsScaledIntegerVertexFormat(format)) { maskOut("native-format-feature", format); continue; }
format = ToFloat32VertexFormat(attr.Size);
conversion = VertexStreamConversion::ScaledIntegerToFloat32;
if (!SupportsVertexBufferFormat(format)) return reject("converted-format-feature", format);
if (!SupportsVertexBufferFormat(format)) { maskOut("converted-format-feature", format); continue; }
}
const SizeT elementSize = GetAttributeByteSize(type, attr.Size, attr.IsBgra);
if (!elementSize) return reject("element-size", format);
if (!elementSize) { maskOut("element-size", format); continue; }
const Uint64 offset = attr.Offset + buffer.Offset;
if (offset < attr.Offset) return reject("offset-overflow", format);
if (offset < attr.Offset) { maskOut("offset-overflow", format); continue; }
const SizeT alignment = (type == DataType::Int2101010Rev || type == DataType::Uint2101010Rev)
? elementSize : GetComponentSize(type);
if (conversion == VertexStreamConversion::None && alignment > 1 &&
@@ -500,8 +553,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// resolved binding/attribute/divisor values, including a legal zero stride.
Uint64 hash = XXH64(out.bindings.data(), out.bindings.size() * sizeof(out.bindings[0]), 0);
hash = XXH64(out.attributes.data(), out.attributes.size() * sizeof(out.attributes[0]), hash);
out.layoutHash = XXH64(out.bindingDivisors.data(),
hash = XXH64(out.bindingDivisors.data(),
out.bindingDivisors.size() * sizeof(out.bindingDivisors[0]), hash);
// THE MASK IS PART OF THE LAYOUT NOW, for the same reason the monolith entry's hash
// carries it (the XXH64_update over unsupportedAttribMask in GetOrCreateVertexInputState
// above). Before this package the mask was always zero here, so leaving it out of the
// hash was free; now two VAOs can produce the SAME bindings, attributes and divisors
// and differ only in which enabled attribute was masked out, and a hash blind to that
// would serve one of them the other's pipeline.
out.layoutHash = XXH64(&out.unsupportedAttribMask, sizeof(out.unsupportedAttribMask), hash);
out.state.vertexBindingDescriptionCount = static_cast<Uint32>(out.bindings.size());
out.state.pVertexBindingDescriptions = out.bindings.data();
out.state.vertexAttributeDescriptionCount = static_cast<Uint32>(out.attributes.size());
@@ -15,12 +15,65 @@
#include "MG_Pipe/MGPipeCallbacks.h"
#include "MG_Pipe/PipeApply.h"
#include "MG_Remote/Server/StagedShadow.h"
#if MOBILEGL_BUILD_DISAGGREGATED
#include "MG_Remote/Transport/AdoptT0.h"
#endif
#if MOBILEGL_BUILD_DISAGGREGATED && defined(__ANDROID__)
#include <android/hardware_buffer.h>
#endif
#include <Config.h>
#include <chrono>
#include <cstdlib>
#endif
namespace MobileGL::MG_Backend::DirectVulkan {
namespace {
#if MOBILEGL_BUILD_DISAGGREGATED
// P7 wave 2 package B3, INVESTIGATION PROBE (split-only, read once, not in
// ConfigLoader's accepted-env table - package B's MGITEST_MAGMA_FORCE_SHADER_MIPMAP
// shape). MOBILEGL_MAGMA_WIREBUF_PROBE=1 prints one line per streamed-buffer event so
// the `glBufferSubData -> draw` ordering of a trace can be read off a log instead of
// inferred. Costs a branch on a cached bool when unset.
Bool WireBufProbeEnabled() {
static const Bool enabled = [] {
const char* value = std::getenv("MOBILEGL_MAGMA_WIREBUF_PROBE");
return value && value[0] == '1';
}();
return enabled;
}
// P7 wave 2 package B3, FORCING KNOB (split-only, read once, not in ConfigLoader's
// accepted-env table - package B's MGITEST_MAGMA_FORCE_SHADER_MIPMAP shape).
//
// WriteWireBuffer has exactly one unsafe shape: taking the immediate host memcpy
// because `lastUseSerial > GetCompletedSerial()` said the buffer is idle. That
// predicate's first term is PURE COUNTING (`m_frameSerial - frameCount`, no fence),
// so on any arm where the frame serial advances without the GPU having proved the
// frame's submissions complete, the answer is a guess. Host lanes never reach that
// state on their own (measured: 0 mid-frame frame-boundary drains over the whole
// OpenRA replay), so the lane cannot arm against it without forcing it.
//
// 0/unset = off. 1 = the serial term answers "idle" for every write it would have
// ordered. N>1 = only the Nth such write (1-based), which is how one
// `glBufferSubData -> draw` pair is poisoned in isolation and the device's "one draw
// contributed nothing" signature is reproduced rather than the whole frame destroyed.
//
// The knob forces ONLY the counting term to lie. It does NOT bypass the ordered copy,
// so it stays a valid red-once across the fix: before the submission term existed the
// lie reached the host memcpy (red); with the submission term the lie is caught and
// the copy is still ordered (green).
Uint64 ForceStaleSerialSelector() {
static const Uint64 selector = [] {
const char* value = std::getenv("MGITEST_MAGMA_FORCE_STALE_BUFFER_SERIAL");
if (!value || !value[0]) return Uint64{0};
return static_cast<Uint64>(std::strtoull(value, nullptr, 10));
}();
return selector;
}
// Counts only the writes the busy predicate wanted to ORDER, so the selector indexes
// the streamed subdata -> draw pairs and nothing else.
Uint64 g_orderedWireWriteCounter = 0;
#endif
constexpr VmaAllocationCreateFlags kResidentBufferAllocationFlags =
VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
constexpr SizeT kLiveResourcePruneThreshold = 256;
@@ -160,6 +213,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// modified ranges. A server pointer is never donated across the wire.
.MapPersistent = [](MG_Pipe::MGPipeHandle, Uint64, const void*) -> void* { return nullptr; },
.UnmapPersistent = [](MG_Pipe::MGPipeHandle) {},
#if defined(__ANDROID__)
// P11 B2 (T0): the one door a server pointer is still never donated through stays shut;
// what T0 adds is the other direction - the CLIENT's pages imported as the store.
.ImportExternal = [](MG_Pipe::MGPipeHandle res, void* ahb, Uint64 size) -> Bool {
return WireManager().ImportWireBuffer(res, ahb, size);
},
.SelfTestExternal = [](char* why, Uint64 whyBytes) -> Bool {
return VkBufferManager::SelfTestWireImport(why, whyBytes);
},
#endif
};
#endif
} // namespace
@@ -194,10 +257,22 @@ namespace MobileGL::MG_Backend::DirectVulkan {
res.Slot, res.Gen);
std::abort();
}
DeferRelease(std::move(resource->buffer));
// THE SERIAL IS READ BEFORE THE RESET BELOW ZEROES IT. Whether it is zero - whether any
// GPU command has named this store - decides whether the orphan can go at once or has
// to wait for a submission (DeferredWireRelease), and three lines from now the record
// no longer carries it.
const Uint64 orphanedUseSerial = resource->lastUseSerial;
// P11 B2: a T0 store is ALWAYS released through the serial-gated list (never "at once"),
// so the client's AHardwareBuffer outlives every submission that can still name it however
// the client's own release (at this respecify) raced the frames in flight.
DeferWireRelease(std::move(resource->buffer),
resource->imported ? T0ReleaseSerial(orphanedUseSerial) : orphanedUseSerial);
resource->imported = false;
resource->size = desc.Width;
resource->lastUseSerial = 0;
resource->lastUseSubmitIndex = 0; // M2 r2: the new store carries no submission yet (B3's stamp is per store)
resource->gpuWritesPending = false;
resource->indirectReadBarrierPending = false;
resource->stagedCoverage.clear();
++m_sliceEpochCounter;
if (resource->size == 0) return;
@@ -216,6 +291,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
res.Slot, res.Gen, static_cast<unsigned long long>(resource->size));
std::abort();
}
++m_wireStoreCount;
NoteWireStorePeaks();
PublishWireReclaimGauges();
// Under transport, defined bytes follow as resource_subdata records. An
// undefined store has no shadow to upload and no implicit zero snapshot.
if (initialBytes != nullptr && desc.HasDefinedContent) {
@@ -224,7 +302,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
Bool VkBufferManager::WaitForWireBufferHostAccess(WireBufferResource& resource) {
if (resource.lastUseSerial <= GetCompletedSerial() && !resource.gpuWritesPending) return true;
// B3 review: the early return must consult the same submission term WriteWireBuffer
// does, or a caller that reaches here with the serial term idle skips the wait the
// term would have demanded.
if (resource.lastUseSerial <= GetCompletedSerial() && !resource.gpuWritesPending &&
(pVulkanRenderer == nullptr || pVulkanRenderer->IsSubmitIndexComplete(resource.lastUseSubmitIndex))) {
return true;
}
if (!m_copyProvider || !pVulkanRenderer) return false;
const VkCommandBuffer commands = m_copyProvider->AcquireBufferCopyCommandBuffer();
if (commands == VK_NULL_HANDLE) return false;
@@ -233,10 +317,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
barrier.dstAccessMask = VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT;
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_HOST_BIT,
0, 1, &barrier, 0, nullptr, 0, nullptr);
if (!pVulkanRenderer->WaitForSubmitIndex(pVulkanRenderer->GetSyncPointSubmitIndex(), UINT64_MAX, true)) {
// P8-D: counted here, past the idle early-out, so `whw` is waits TAKEN, not calls.
++m_wireHostWaits;
const Bool timed = MG_Util::PipeStats::Enabled();
const auto waitStart = timed ? std::chrono::steady_clock::now() : std::chrono::steady_clock::time_point{};
const Bool waited =
pVulkanRenderer->WaitForSubmitIndex(pVulkanRenderer->GetSyncPointSubmitIndex(), UINT64_MAX, true);
if (timed) {
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireHostWaits, 1);
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireHostWaitMicros,
static_cast<Uint64>(std::chrono::duration_cast<std::chrono::microseconds>(
std::chrono::steady_clock::now() - waitStart).count()));
}
if (!waited) {
return false;
}
resource.lastUseSerial = 0;
// B3: the wait above proved every submission at or below the sync point complete, so
// the submission term must be cleared with the serial or it would keep answering busy.
resource.lastUseSubmitIndex = 0;
resource.gpuWritesPending = false;
return true;
}
@@ -251,7 +350,52 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
if (size == 0) return;
Bool uploaded = false;
if (resource->lastUseSerial > GetCompletedSerial()) {
const Uint64 probeLastUse = resource->lastUseSerial;
const Uint64 probeCompleted = GetCompletedSerial();
// ---- P7 wave 2 package B3: WHAT EACH TERM OF THIS PREDICATE IS FOR -----------------
//
// A streamed `glBufferSubData -> draw` pair is only correct if the write is ordered
// against every draw that still reads the bytes it overwrites; a write this predicate
// answers "idle" is applied as an unsynchronised host memcpy, and a draw still reading
// the range then renders bytes that are not its own - silently, with no log line, no GL
// error and no Fatal. That was the OpenRA device divergence (magma-b3.md §2).
//
// THE GUARANTEE IS THE SERIAL TERM, and it is sound because of the FLOOR, not because
// of anything here. GetCompletedSerial() is max(m_frameSerial - frameCount,
// m_completedSerialFloor). The floor used to be raised to a retired submission's frame
// serial even while another submission carrying the SAME serial was still executing -
// on the wire arm a serial has two (a mid-frame FlushPendingCommands under a pooled
// fence, then Present) - so "frame N-1 complete" was asserted on a fence nobody waited.
// VulkanRenderer::OnSubmitsCompletedUpTo now raises it only to a serial no in-flight
// submission still carries, and run_trace_case.cmake reds a split retrace on any
// "MGWIRE-FLOOR unsound-serial-complete" line, which is that fix's lane.
//
// THE SUBMISSION TERM IS A PROBE AND A DEFENCE, NOT A SECOND GUARANTEE. It is not sound
// on its own: lastUseSubmitIndex is stamped at AcquireWireSlice with the next submission
// index, but the draw that acquired the slice can still be pushed into a LATER
// submission - BindProgramUniformBuffers -> SyncWireTextureShape's preserve path ->
// FlushWirePendingCommandsForTextureUpdate submits the stamped index without the draw -
// so the stamp can name S1 while the draw rides S2. What it does do: with the floor
// sound it should never be the term that says "busy" (the WBUF probe reports bySubmit
// separately, so a device that shows one has found a floor hole), and it is the only
// thing the MGITEST_MAGMA_FORCE_STALE_BUFFER_SERIAL knob leaves standing, which is what
// the StaleSerial. entries pin.
Bool busyBySerial = resource->lastUseSerial > GetCompletedSerial();
if (busyBySerial && ForceStaleSerialSelector() != 0) {
const Uint64 index = ++g_orderedWireWriteCounter;
const Uint64 selector = ForceStaleSerialSelector();
if (selector == 1 || selector == index) {
busyBySerial = false;
MGLOG_W("MGITEST_MAGMA_FORCE_STALE_BUFFER_SERIAL: write #%llu {slot=%u, gen=%u} off=%llu "
"size=%llu - the counting term is forced to answer \"idle\"",
static_cast<unsigned long long>(index), res.Slot, res.Gen,
static_cast<unsigned long long>(offset), static_cast<unsigned long long>(size));
}
}
const Bool busyBySubmission =
pVulkanRenderer != nullptr && !pVulkanRenderer->IsSubmitIndexComplete(resource->lastUseSubmitIndex);
if (busyBySerial || busyBySubmission) {
// Vulkan buffer copies require four-byte aligned ranges. An odd GL
// byte update waits, then writes only the exact range; rounding it
// from a CPU shadow could overwrite neighbouring GPU-written bytes.
@@ -259,10 +403,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
uploaded = StagedWireRangeCopy(*resource, bytes, static_cast<SizeT>(offset),
static_cast<SizeT>(size));
}
if (!uploaded && !WaitForWireBufferHostAccess(*resource)) {
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-write-sync\"}");
std::abort();
}
if (!uploaded && !WaitForWireBufferHostAccess(*resource)) WireBufferSyncFatal("host-write");
}
if (!uploaded) {
uploaded = resource->buffer.Upload(bytes, size, offset);
@@ -274,6 +415,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-upload\"}");
std::abort();
}
if (WireBufProbeEnabled()) {
MGLOG_I("WBUF write slot=%u gen=%u off=%llu size=%llu branch=%s bySerial=%d bySubmit=%d "
"lastUse=%llu completed=%llu frameSerial=%llu floor=%llu lastUseSubmit=%llu",
res.Slot, res.Gen, static_cast<unsigned long long>(offset),
static_cast<unsigned long long>(size),
(busyBySerial || busyBySubmission) ? "STAGED-or-WAIT" : "HOST-IMMEDIATE",
busyBySerial ? 1 : 0, busyBySubmission ? 1 : 0,
static_cast<unsigned long long>(probeLastUse), static_cast<unsigned long long>(probeCompleted),
static_cast<unsigned long long>(m_frameSerial),
static_cast<unsigned long long>(m_completedSerialFloor),
static_cast<unsigned long long>(resource->lastUseSubmitIndex));
}
MG_Remote::Server::StagedShadowStore::CoverageAdd(resource->stagedCoverage,
static_cast<SizeT>(offset),
static_cast<SizeT>(offset + size));
@@ -304,8 +457,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
auto* resource = FindWireBuffer(res);
if (!resource || !resource->buffer.IsValid() || resource->size == 0) return false;
resource->lastUseSerial = m_frameSerial;
// B3: the draw this slice is being acquired for is normally recorded into the NEXT
// submission - normally, not always: a mid-draw flush (SyncWireTextureShape's preserve
// path) can submit this index without the draw. This stamp is the defence term's input
// (see WriteWireBuffer); the floor behind lastUseSerial is the guarantee.
if (pVulkanRenderer) resource->lastUseSubmitIndex = pVulkanRenderer->GetWireNextSubmitIndex();
outSlice = resource->buffer.GetSlice(0, resource->size);
outSlice.mapped = nullptr;
if (WireBufProbeEnabled()) {
MGLOG_I("WBUF bind slot=%u gen=%u frameSerial=%llu", res.Slot, res.Gen,
static_cast<unsigned long long>(m_frameSerial));
}
return true;
}
@@ -323,6 +485,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// work, not the draw that will bind that previously acquired slice.
// Preserve its reservation even if the idle drain advanced the frame.
resource->lastUseSerial = m_frameSerial;
// B3 review: and the submission half of the same reservation - the wait cleared it,
// and the draw holding the slice has not been submitted yet.
if (pVulkanRenderer) resource->lastUseSubmitIndex = pVulkanRenderer->GetWireNextSubmitIndex();
}
if (!resource->buffer.Invalidate(size, offset)) return false;
Memcpy(dst, static_cast<const Uint8*>(resource->buffer.GetMappedData()) + offset, static_cast<SizeT>(size));
@@ -353,12 +518,77 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return true;
}
Bool VkBufferManager::CopyWireBufferSubWordRangeToSlice(MG_Pipe::MGPipeHandle res, Uint64 offset,
Uint64 size, Uint32 frameIndex,
const BufferSlice& dst, Uint64 dstSkip) {
auto* resource = FindWireBuffer(res);
if (!resource || offset > resource->size || size > resource->size - offset) return false;
if (dstSkip > dst.size || size > dst.size - dstSkip) return false;
if (size == 0) return true;
if (!m_copyProvider || dst.buffer == VK_NULL_HANDLE) return false;
// Round the read of the application's store OUT to whole words. The tail can only be
// clipped by the end of the store itself, and [offset, offset + size) already fits
// inside it, so the widened window still covers every byte the caller asked for.
const Uint64 alignedOffset = offset & ~Uint64{3};
const Uint64 stagedEnd = std::min<Uint64>((offset + size + 3) & ~Uint64{3}, resource->size);
const Uint64 headPad = offset - alignedOffset;
const Uint64 stagedSize = stagedEnd - alignedOffset;
if (stagedSize < headPad + size) return false;
// Uninitialized on purpose: the widening copy below writes every byte of it. The
// arena can grow to satisfy this, which parks (never frees) the buffer `dst` may
// already name - see BufferArena::EnsureCapacity - so the two slices are allowed to
// sit in different VkBuffers and the copy names each slice's own handle.
BufferSlice staging{};
if (!m_transientUploadArena.Allocate(frameIndex, stagedSize, 4, staging) || !staging.IsValid())
return false;
const VkCommandBuffer commands = m_copyProvider->AcquireBufferCopyCommandBuffer();
if (commands == VK_NULL_HANDLE) return false;
VkMemoryBarrier before{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
before.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_HOST_WRITE_BIT;
before.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT | VK_PIPELINE_STAGE_HOST_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1, &before, 0, nullptr, 0, nullptr);
const VkBufferCopy widen{alignedOffset, staging.offset, stagedSize};
vkCmdCopyBuffer(commands, resource->buffer.GetHandle(), staging.buffer, 1, &widen);
// The shift reads what the widening copy just wrote, so it needs its own edge even
// though both halves are transfers on one command buffer.
VkMemoryBarrier staged{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
staged.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
staged.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
0, 1, &staged, 0, nullptr, 0, nullptr);
const VkBufferCopy shift{staging.offset + headPad, dst.offset + dstSkip, size};
vkCmdCopyBuffer(commands, staging.buffer, dst.buffer, 1, &shift);
VkMemoryBarrier after{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
after.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
after.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, 1, &after, 0, nullptr, 0, nullptr);
resource->lastUseSerial = m_frameSerial;
return true;
}
void VkBufferManager::ReadbackWireBuffer(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size) {
Vector<Uint8> bytes(static_cast<SizeT>(size));
if (!ReadWireBuffer(res, offset, size, bytes.data())) {
// P11 B2: A T0 STORE'S READBACK IS "DONE, NO BYTES". Its pages are the client's
// AHardwareBuffer, so what the client needs before it reads them is this server's GPU
// having finished every write to them - the resident copies WriteWireBuffer recorded and
// any shader write - and nothing crossing SEG_EVENT. The wait is the store's own host-access
// wait (a barrier to host, then the sync point's submission); the reply then says "done".
// One death for both shapes: the store could not be made readable.
auto* resource = FindWireBuffer(res);
const Bool t0 = resource != nullptr && resource->imported;
Vector<Uint8> bytes(t0 ? 0 : static_cast<SizeT>(size));
if (t0 ? !WaitForWireBufferHostAccess(*resource) : !ReadWireBuffer(res, offset, size, bytes.data())) {
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-readback\"} {slot=%u, gen=%u}", res.Slot, res.Gen);
std::abort();
}
if (t0) {
++m_sliceEpochCounter;
return;
}
if (MG_Pipe::gMGPipeCallbacks.OnBufferWriteback == nullptr) {
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-writeback-callback\"}");
std::abort();
@@ -378,6 +608,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
resource->lastUseSerial = m_frameSerial;
resource->stagedCoverage.clear();
resource->gpuWritesPending = true;
resource->indirectReadBarrierPending = true;
++m_sliceEpochCounter;
// The existing client dirty-state protocol accepts one whole-store
// notification. The store remains on the GPU and exact later SubData
@@ -390,12 +621,439 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MG_Pipe::gMGPipeCallbacks.OnGpuWritten(res, 1, &range);
}
Bool VkBufferManager::TakeWireIndirectReadBarrier(MG_Pipe::MGPipeHandle res) {
auto* resource = FindWireBuffer(res);
if (!resource || !resource->indirectReadBarrierPending) return false;
resource->indirectReadBarrierPending = false;
return true;
}
// ---- P11 B2: T0 - THE CLIENT'S AHardwareBuffer AS A WIRE STORE -----------------------------
// The serial a T0 store is released under: its last use, or - when no GPU command has named it
// since a host-access wait - the current frame's, so the release still waits for the sync
// point rather than taking DeferWireRelease's "never named, destroy now" arm. Conservative by
// at most one submission.
Uint64 VkBufferManager::T0ReleaseSerial(Uint64 lastUseSerial) const {
if (lastUseSerial != 0) return lastUseSerial;
return m_frameSerial != 0 ? m_frameSerial : 1;
}
#if defined(__ANDROID__)
namespace {
Int32 T0PickMemoryType(const VkPhysicalDeviceMemoryProperties& props, Uint32 bits,
VkMemoryPropertyFlags want) {
for (Uint32 i = 0; i < props.memoryTypeCount; ++i) {
if ((bits & (1u << i)) != 0 && (props.memoryTypes[i].propertyFlags & want) == want)
return static_cast<Int32>(i);
}
return -1;
}
struct T0ImportedBuffer {
VkBuffer buffer = VK_NULL_HANDLE;
VkDeviceMemory memory = VK_NULL_HANDLE;
void* mapped = nullptr;
Bool coherent = true;
};
void T0DestroyImported(VkDevice device, T0ImportedBuffer& imported) {
if (imported.mapped != nullptr) vkUnmapMemory(device, imported.memory);
if (imported.buffer != VK_NULL_HANDLE) vkDestroyBuffer(device, imported.buffer, nullptr);
if (imported.memory != VK_NULL_HANDLE) vkFreeMemory(device, imported.memory, nullptr);
imported = T0ImportedBuffer{};
}
// A VkBuffer on a dedicated allocation IMPORTED from `ahb`, in a host-visible memory type
// (coherent preferred), mapped whole. The server's own host map is what WriteWireBuffer's
// idle path and every CPU reader of a wire store use, so a type without HOST_VISIBLE is a
// refusal, not a degraded import. Nothing is left behind on failure.
Bool T0ImportAhb(const VulkanRenderer::WireAhbImport& ctx, void* ahb, Uint64 size, VkBufferUsageFlags usage,
T0ImportedBuffer& out, String& why) {
out = T0ImportedBuffer{};
const auto getProperties =
reinterpret_cast<PFN_vkGetAndroidHardwareBufferPropertiesANDROID>(ctx.getAhbProperties);
VkAndroidHardwareBufferPropertiesANDROID properties{
VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID};
VkResult result = getProperties(ctx.device, static_cast<const AHardwareBuffer*>(ahb), &properties);
if (result != VK_SUCCESS) {
why = "vkGetAndroidHardwareBufferPropertiesANDROID -> " + std::to_string(static_cast<int>(result));
return false;
}
VkExternalMemoryBufferCreateInfo external{VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO};
external.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID;
VkBufferCreateInfo bufferInfo{VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO};
bufferInfo.pNext = &external;
bufferInfo.size = size;
bufferInfo.usage = usage;
bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
result = vkCreateBuffer(ctx.device, &bufferInfo, nullptr, &out.buffer);
if (result != VK_SUCCESS) {
out.buffer = VK_NULL_HANDLE;
why = "vkCreateBuffer(external AHB) -> " + std::to_string(static_cast<int>(result));
return false;
}
Int32 type = T0PickMemoryType(ctx.memory, properties.memoryTypeBits,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
out.coherent = type >= 0;
if (type < 0)
type = T0PickMemoryType(ctx.memory, properties.memoryTypeBits, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT);
if (type < 0) {
why = "no host-visible memory type can import it (memoryTypeBits 0x" +
std::to_string(properties.memoryTypeBits) + ")";
T0DestroyImported(ctx.device, out);
return false;
}
VkMemoryDedicatedAllocateInfo dedicated{VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO};
dedicated.buffer = out.buffer;
VkImportAndroidHardwareBufferInfoANDROID import{VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID};
import.buffer = static_cast<AHardwareBuffer*>(ahb);
import.pNext = &dedicated;
VkMemoryAllocateInfo allocateInfo{VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO};
allocateInfo.pNext = &import;
allocateInfo.allocationSize = properties.allocationSize;
allocateInfo.memoryTypeIndex = static_cast<Uint32>(type);
result = vkAllocateMemory(ctx.device, &allocateInfo, nullptr, &out.memory);
if (result != VK_SUCCESS) {
out.memory = VK_NULL_HANDLE;
why = "vkAllocateMemory(import AHB) -> " + std::to_string(static_cast<int>(result));
T0DestroyImported(ctx.device, out);
return false;
}
result = vkBindBufferMemory(ctx.device, out.buffer, out.memory, 0);
if (result == VK_SUCCESS) result = vkMapMemory(ctx.device, out.memory, 0, VK_WHOLE_SIZE, 0, &out.mapped);
if (result != VK_SUCCESS || out.mapped == nullptr) {
out.mapped = nullptr;
why = "vkBindBufferMemory/vkMapMemory -> " + std::to_string(static_cast<int>(result));
T0DestroyImported(ctx.device, out);
return false;
}
return true;
}
} // namespace
#endif
Bool VkBufferManager::ImportWireBuffer(MG_Pipe::MGPipeHandle res, void* ahb, Uint64 size) {
#if defined(__ANDROID__)
auto* resource = FindWireBuffer(res);
if (resource == nullptr || ahb == nullptr || size == 0 || size != resource->size) return false;
VulkanRenderer::WireAhbImport ctx;
if (pVulkanRenderer == nullptr || !pVulkanRenderer->GetWireAhbImport(ctx)) {
MGLOG_W_ONCE("Magma T0: this device did not take VK_ANDROID_external_memory_android_hardware_buffer");
return false;
}
VkBufferUsageFlags usage = kPersistentBackedUsage | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
if (m_initInfo.transformFeedbackUsageEnabled) usage |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
T0ImportedBuffer imported;
String why;
if (!T0ImportAhb(ctx, ahb, size, usage, imported, why)) {
MGLOG_W("Magma T0: store {slot=%u, gen=%u} (%llu bytes) not imported - %s", res.Slot, res.Gen,
static_cast<unsigned long long>(size), why.c_str());
return false;
}
// The T2 store it replaces goes the way a respecify's does (and any draw still naming it
// keeps it until its submission completes).
DeferWireRelease(std::move(resource->buffer), resource->lastUseSerial);
MG_Remote::Transport::AdoptT0::AcquireImported(ahb);
resource->buffer.AdoptExternal(ctx.device, imported.buffer, imported.memory, size, imported.mapped,
imported.coherent, ahb);
resource->imported = true;
resource->lastUseSerial = 0;
resource->lastUseSubmitIndex = 0;
resource->gpuWritesPending = false;
// The whole store holds bytes the client supplied (its AHB, seeded from its shadow).
resource->stagedCoverage.clear();
MG_Remote::Server::StagedShadowStore::CoverageAdd(resource->stagedCoverage, 0, static_cast<SizeT>(size));
++m_sliceEpochCounter;
++m_wireStoreCount;
NoteWireStorePeaks();
PublishWireReclaimGauges();
MGLOG_D("Magma T0: imported {slot=%u, gen=%u} (%llu bytes, %s memory)", res.Slot, res.Gen,
static_cast<unsigned long long>(size), imported.coherent ? "coherent" : "non-coherent");
return true;
#else
(void)res;
(void)ahb;
(void)size;
return false;
#endif
}
// THE POST (B2; decided by probe, never by driver name). The AHB API promises CPU coherence only
// across an unlock/lock, and T0 holds the client's lock for the store's life, so this exercises
// that pattern on THIS device before kCapAdoptT0 may be published:
// hold - a 64 KiB BLOB allocated and locked by the test, never unlocked while it runs;
// read - this server's own host map of the import reads what the CPU wrote through the lock;
// gpu - one submission copies that pattern out of the import (the GPU read) and fills another
// region of it (the GPU write), then a transfer->host barrier and a fence wait;
// back - the fill is read through the HELD pointer, the copy from a host-visible staging buffer.
Bool VkBufferManager::SelfTestWireImport(char* whyOut, Uint64 whyBytes) {
const auto answer = [&](Bool ok, const String& text) {
std::snprintf(whyOut, static_cast<SizeT>(whyBytes), "%s", text.c_str());
return ok;
};
#if defined(__ANDROID__)
VulkanRenderer::WireAhbImport ctx;
if (pVulkanRenderer == nullptr || !pVulkanRenderer->GetWireAhbImport(ctx))
return answer(false, "the device did not take VK_ANDROID_external_memory_android_hardware_buffer");
constexpr Uint64 kBytes = 64 * 1024;
constexpr Uint32 kWords = 1024; // one 4 KiB region
constexpr Uint64 kFillAt = 4 * 4096; // region 4
constexpr Uint32 kFillWord = 0x6C0FFEE0u;
Vector<Uint32> pattern(kBytes / 4);
for (SizeT i = 0; i < pattern.size(); ++i) pattern[i] = 0x5A000000u ^ static_cast<Uint32>(i * 2654435761u);
MG_Remote::Transport::AdoptT0::HeldStore held;
String why;
if (!MG_Remote::Transport::AdoptT0::AllocateHeld(kBytes, pattern.data(), held, why))
return answer(false, "hold: " + why);
T0ImportedBuffer imported, staging;
VkCommandPool pool = VK_NULL_HANDLE;
VkFence fence = VK_NULL_HANDLE;
String verdict;
Bool ok = false;
do {
if (!T0ImportAhb(ctx, held.ahb, kBytes,
VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
imported, why)) {
verdict = "import: " + why;
break;
}
if (!imported.coherent) {
VkMappedMemoryRange range{VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE};
range.memory = imported.memory;
range.size = VK_WHOLE_SIZE;
vkInvalidateMappedMemoryRanges(ctx.device, 1, &range);
}
if (std::memcmp(imported.mapped, pattern.data(), kWords * 4) != 0) {
verdict = "read: the server's host map of the import does not hold the CPU's pattern";
break;
}
// A host-visible staging buffer for the GPU read's destination.
VkBufferCreateInfo stagingInfo{VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO};
stagingInfo.size = kWords * 4;
stagingInfo.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT;
stagingInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
if (vkCreateBuffer(ctx.device, &stagingInfo, nullptr, &staging.buffer) != VK_SUCCESS) {
staging.buffer = VK_NULL_HANDLE;
verdict = "staging buffer";
break;
}
VkMemoryRequirements requirements{};
vkGetBufferMemoryRequirements(ctx.device, staging.buffer, &requirements);
const Int32 stagingType = T0PickMemoryType(ctx.memory, requirements.memoryTypeBits,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
VkMemoryAllocateInfo stagingAlloc{VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO};
stagingAlloc.allocationSize = requirements.size;
stagingAlloc.memoryTypeIndex = static_cast<Uint32>(stagingType < 0 ? 0 : stagingType);
if (stagingType < 0 || vkAllocateMemory(ctx.device, &stagingAlloc, nullptr, &staging.memory) != VK_SUCCESS ||
vkBindBufferMemory(ctx.device, staging.buffer, staging.memory, 0) != VK_SUCCESS ||
vkMapMemory(ctx.device, staging.memory, 0, VK_WHOLE_SIZE, 0, &staging.mapped) != VK_SUCCESS) {
verdict = "staging memory";
break;
}
std::memset(staging.mapped, 0, kWords * 4);
VkCommandPoolCreateInfo poolInfo{VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO};
poolInfo.queueFamilyIndex = ctx.queueFamily;
poolInfo.flags = VK_COMMAND_POOL_CREATE_TRANSIENT_BIT;
VkCommandBuffer commands = VK_NULL_HANDLE;
VkCommandBufferAllocateInfo commandInfo{VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO};
commandInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
commandInfo.commandBufferCount = 1;
VkFenceCreateInfo fenceInfo{VK_STRUCTURE_TYPE_FENCE_CREATE_INFO};
if (vkCreateCommandPool(ctx.device, &poolInfo, nullptr, &pool) != VK_SUCCESS) {
pool = VK_NULL_HANDLE;
verdict = "command pool";
break;
}
commandInfo.commandPool = pool;
if (vkAllocateCommandBuffers(ctx.device, &commandInfo, &commands) != VK_SUCCESS ||
vkCreateFence(ctx.device, &fenceInfo, nullptr, &fence) != VK_SUCCESS) {
verdict = "command buffer / fence";
break;
}
VkCommandBufferBeginInfo begin{VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO};
begin.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
vkBeginCommandBuffer(commands, &begin);
VkMemoryBarrier hostWrites{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
hostWrites.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT;
hostWrites.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT;
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 1,
&hostWrites, 0, nullptr, 0, nullptr);
VkBufferCopy region{0, 0, kWords * 4};
vkCmdCopyBuffer(commands, imported.buffer, staging.buffer, 1, &region);
vkCmdFillBuffer(commands, imported.buffer, kFillAt, kWords * 4, kFillWord);
VkMemoryBarrier toHost{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
toHost.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
toHost.dstAccessMask = VK_ACCESS_HOST_READ_BIT;
vkCmdPipelineBarrier(commands, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0, 1, &toHost,
0, nullptr, 0, nullptr);
vkEndCommandBuffer(commands);
VkSubmitInfo submit{VK_STRUCTURE_TYPE_SUBMIT_INFO};
submit.commandBufferCount = 1;
submit.pCommandBuffers = &commands;
if (vkQueueSubmit(ctx.queue, 1, &submit, fence) != VK_SUCCESS ||
vkWaitForFences(ctx.device, 1, &fence, VK_TRUE, 2'000'000'000ull) != VK_SUCCESS) {
verdict = "gpu: the submission did not complete within 2 s";
break;
}
Uint32 gpuRead = 0, heldSaw = 0;
const auto* copied = static_cast<const Uint32*>(staging.mapped);
const auto* heldWords = static_cast<const Uint32*>(held.ptr);
for (Uint32 i = 0; i < kWords; ++i) {
if (copied[i] == pattern[i]) ++gpuRead;
if (heldWords[kFillAt / 4 + i] == kFillWord) ++heldSaw;
}
ok = gpuRead == kWords && heldSaw == kWords;
verdict = "64 KiB AHB, lock held throughout: host map read ok (" +
String(imported.coherent ? "coherent" : "non-coherent") + " memory), GPU read " +
std::to_string(gpuRead) + "/1024 words of the CPU's pattern, GPU write seen through the "
"held lock " + std::to_string(heldSaw) + "/1024 words";
} while (false);
if (fence != VK_NULL_HANDLE) vkDestroyFence(ctx.device, fence, nullptr);
if (pool != VK_NULL_HANDLE) vkDestroyCommandPool(ctx.device, pool, nullptr);
T0DestroyImported(ctx.device, staging);
T0DestroyImported(ctx.device, imported);
MG_Remote::Transport::AdoptT0::ReleaseHeld(held);
return answer(ok, verdict);
#else
return answer(false, "no AHardwareBuffer on this platform");
#endif
}
void VkBufferManager::DestroyWireBuffer(MG_Pipe::MGPipeHandle res) {
const auto found = m_wireBuffers.find(WireBufferKey(res));
if (found == m_wireBuffers.end()) return;
DeferRelease(std::move(found->second.buffer));
DeferWireRelease(std::move(found->second.buffer),
found->second.imported ? T0ReleaseSerial(found->second.lastUseSerial)
: found->second.lastUseSerial); // P11 B2, as Respecify
m_wireBuffers.erase(found);
++m_sliceEpochCounter;
PublishWireReclaimGauges();
}
void VkBufferManager::DeferWireRelease(VkBufferObject&& buffer, Uint64 lastUseSerial) {
if (!buffer.IsValid()) return;
// Two stores go NOW rather than into the list. With no frame slots (DeferRelease's own
// teardown guard, and for its reason) there is no manager left to reclaim anything
// later. And a store whose last-use serial is 0 was never named by a GPU command since
// it was minted or since a host-access wait retired everything recorded before it - see
// DeferredWireRelease - so there is nothing to wait for.
if (m_deferredBufferReleases.empty() || lastUseSerial == 0) {
buffer.Destroy();
++m_wireStoreDestroyEpoch;
if (m_wireStoreCount > 0) --m_wireStoreCount;
return;
}
DeferredWireRelease entry;
entry.lastUseSerial = lastUseSerial;
// The last submission that can name this store: see DeferredWireRelease. Zero when there
// is no renderer, which is also when there is no queue and nothing to wait for.
entry.submitIndex = pVulkanRenderer != nullptr ? pVulkanRenderer->GetSyncPointSubmitIndex() : 0;
entry.bytes = static_cast<Uint64>(buffer.GetSize());
entry.buffer = std::move(buffer);
m_deferredWireBytes += entry.bytes;
m_deferredWireReleases.push_back(std::move(entry));
NoteWireStorePeaks();
// EVERY PARK SWEEPS. The frame boundary is not a reclaim point this arm can lean on - a
// pbuffer replay that snapshot-exits delivers ONE present for 1.3 M calls (ID-P7-32) -
// so the only cadence that tracks the workload is the workload itself.
SweepDeferredWireReleases();
EnforceWireDeferredWatermark();
}
void VkBufferManager::EnforceWireDeferredWatermark() {
const Uint64 budget = static_cast<Uint64>(MG_Config::Ipc.WireDeferredMb) * 1024u * 1024u;
// TWO TRIGGERS, ONE SWITCH. Bytes are what the knob names, but a VkBuffer costs per
// OBJECT as well as per byte, and small orphans never reach a byte budget: measured on
// bsl-esc-menu-854 (spawn, lavapipe), one stretch with no submission parked 12,498 stores
// in 39.5 MB against 64 MiB. So a fixed count ceiling rides the same sync point; the
// knob's 0 turns both off, which keeps it one negative control.
if (budget == 0) return;
if (m_deferredWireBytes <= budget && m_deferredWireReleases.size() <= kWireDeferredCountCeiling) return;
// What is left after the sweep is named by work that is recorded but unsubmitted, or
// submitted but unretired. Every entry is tagged at or below the sync point taken HERE,
// so waiting it out - the host-access wait's own call, which flushes the recording first
// - proves all of them dead. A mid-frame flush is what WaitForWireBufferHostAccess and a
// glClientWaitSync already do to this command stream; it costs one submission per
// budget's worth of orphans, not one per glBufferData.
if (pVulkanRenderer == nullptr ||
!pVulkanRenderer->WaitForSubmitIndex(pVulkanRenderer->GetSyncPointSubmitIndex(), UINT64_MAX, true)) {
WireBufferSyncFatal("deferred-watermark");
}
++m_wireDeferredSyncs;
SweepDeferredWireReleases();
}
void VkBufferManager::NoteWireStorePeaks() {
m_wireStoreCountPeak = std::max(m_wireStoreCountPeak, m_wireStoreCount);
m_deferredWireBytesPeak = std::max(m_deferredWireBytesPeak, m_deferredWireBytes);
}
void VkBufferManager::PublishWireReclaimGauges() {
if (!MG_Util::PipeStats::Enabled()) return;
using MG_Util::PipeStats::Gauge;
MG_Util::PipeStats::PublishGauge(Gauge::WireBuffers, static_cast<Uint64>(m_wireBuffers.size()));
MG_Util::PipeStats::PublishGauge(Gauge::WireStoresPeak, m_wireStoreCountPeak);
MG_Util::PipeStats::PublishGauge(Gauge::WireDeferredBytesPeak, m_deferredWireBytesPeak);
MG_Util::PipeStats::PublishGauge(Gauge::WireDeferredSyncs, m_wireDeferredSyncs);
}
void VkBufferManager::WireBufferSyncFatal(const char* site) {
MGLOG_F("Magma: Fatal{ResourceUnavailable, \"buffer-write-sync\"} {site=%s}", site);
std::abort();
}
SizeT VkBufferManager::SweepDeferredWireReleases() {
if (m_deferredWireReleases.empty()) return 0;
SizeT dead = 0;
if (pVulkanRenderer == nullptr ||
pVulkanRenderer->IsSubmitIndexComplete(pVulkanRenderer->GetSyncPointSubmitIndex())) {
// Idle: nothing recorded, every submission retired - or no renderer, so no queue.
// Every parked store is dead, including one tagged for a batch that was abandoned
// instead of submitted.
dead = m_deferredWireReleases.size();
} else {
// The list is in park order and GetSyncPointSubmitIndex() steps back only when a
// pending recording is abandoned instead of submitted (RecreateSwapchain and the
// minimized Present force-clear the recording flags, so `m_submitCounter + 1`
// becomes `m_submitCounter`). Everywhere else the submit indices are non-decreasing
// and the dead entries are a PREFIX: the first entry the renderer will not call
// complete proves none after it is either. Across an abandonment a later entry can
// carry the LOWER index; the walk then stops at the earlier, higher one and holds
// both until the next submission takes that index and retires - conservative, never
// early, and the idle rule above retires them regardless. That bounds the walk to a
// couple of fence polls, which matters because this runs on every glBufferData the
// wire carries.
while (dead < m_deferredWireReleases.size() &&
pVulkanRenderer->IsSubmitIndexComplete(m_deferredWireReleases[dead].submitIndex)) {
++dead;
}
}
if (dead == 0) return 0;
for (SizeT i = 0; i < dead; ++i) {
DeferredWireRelease& entry = m_deferredWireReleases[i];
m_deferredWireBytes -= entry.bytes;
entry.buffer.Destroy();
}
// One bump per sweep that destroyed anything is enough: what a memo needs to know
// is "some handle it may name has been freed since", not how many.
++m_wireStoreDestroyEpoch;
m_deferredWireReleases.erase(m_deferredWireReleases.begin(),
m_deferredWireReleases.begin() + static_cast<std::ptrdiff_t>(dead));
const Uint64 destroyedCount = static_cast<Uint64>(dead);
m_wireStoreCount = destroyedCount <= m_wireStoreCount ? m_wireStoreCount - destroyedCount : 0;
return dead;
}
void VkBufferManager::DestroyAllDeferredWireReleases() {
if (!m_deferredWireReleases.empty()) ++m_wireStoreDestroyEpoch;
for (auto& entry : m_deferredWireReleases) {
entry.buffer.Destroy();
if (m_wireStoreCount > 0) --m_wireStoreCount;
}
m_deferredWireReleases.clear();
m_deferredWireBytes = 0;
}
#endif
@@ -440,7 +1098,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
DestroyAllDeferredReleases();
ReleaseAllLiveResources();
#if MOBILEGL_BUILD_DISAGGREGATED
// DestroyAllDeferredReleases above emptied the parked list; this destroys the stores the
// records still hold, so nothing this arm minted outlives the count.
if (!m_wireBuffers.empty()) ++m_wireStoreDestroyEpoch;
m_wireBuffers.clear();
m_wireStoreCount = 0;
m_wireStoreCountPeak = 0;
m_deferredWireBytesPeak = 0;
m_wireDeferredSyncs = 0;
#endif
m_copyProvider = nullptr;
m_initInfo = {};
@@ -470,6 +1135,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
"VkBufferManager::BeginFrame frame index out of range");
m_currentFrameIndex = frameIndex;
++m_frameSerial;
#if MOBILEGL_BUILD_DISAGGREGATED
if (WireBufProbeEnabled()) {
MGLOG_I("WBUF BEGINFRAME idx=%u newSerial=%llu floor=%llu (arena slot rewound)", frameIndex,
static_cast<unsigned long long>(m_frameSerial),
static_cast<unsigned long long>(m_completedSerialFloor));
}
#endif
CollectDeferredReleases(frameIndex);
m_transientUploadArena.BeginFrame(frameIndex);
}
@@ -486,8 +1158,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Nothing bumps the slice epoch for those other resources, so freeing the buffer
// here left the streamed memo handing a destroyed VkBuffer to vkCmdBindIndexBuffer
// (llvmpipe then faulted inside the draw; the Create/Flywheel indirect retrace died
// exactly this way). Mid-frame drains do not advance m_frameSerial, so they must not
// free arena storage either: the arena's own ResetFrame/BeginFrame is the point where
// exactly this way). Most mid-frame drains do not advance m_frameSerial; every
// eighth does, but only through BeginFrame after the arena/memo boundary work.
// The drain's per-resource sweep must not free arena storage: ResetFrame/BeginFrame is where
// the slot's slices stop being reachable, and that is where these releases land.
for (Uint32 frameIndex = 0; frameIndex < m_deferredBufferReleases.size(); ++frameIndex) {
CollectDeferredReleases(frameIndex);
@@ -766,6 +1439,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
&afterBarrier, 0, nullptr, 0, nullptr);
resource.lastUseSerial = m_frameSerial;
// B3: the copy just recorded rides the next submission too, so a later host write to
// the same bytes must wait for it exactly as a draw's read must.
if (pVulkanRenderer) resource.lastUseSubmitIndex = pVulkanRenderer->GetWireNextSubmitIndex();
resource.gpuWritesPending = true;
return true;
}
@@ -1147,6 +1823,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
"VkBufferManager::CollectDeferredReleases frame index out of range");
m_deferredBufferReleases[frameIndex].clear();
m_deferredResourceReleases[frameIndex].clear();
#if MOBILEGL_BUILD_DISAGGREGATED
// The wire list is not per-slot and does not wait for a frame boundary; a boundary is
// just one more point to sweep at (BeginFrame after its slot fence, and every slot of
// the idle drain, where the renderer-idle rule empties the list).
SweepDeferredWireReleases();
#endif
}
BufferSlice VkBufferManager::AcquireUnboundStorageDescriptor() {
@@ -1242,6 +1924,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
void VkBufferManager::DestroyAllDeferredReleases() {
#if MOBILEGL_BUILD_DISAGGREGATED
// Both callers (Shutdown, RecreateTransientArenas) have proven the device idle.
DestroyAllDeferredWireReleases();
#endif
for (auto& releases : m_deferredBufferReleases) {
for (auto& buffer : releases) {
buffer.Destroy();
@@ -113,7 +113,26 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool ReadWireBuffer(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size, void* dst);
Bool CopyWireBufferRangeToSlice(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size,
const BufferSlice& dst);
// P7 A.1, the sub-word half of the one above: the same copy for a window whose start
// or end does not land on a four-byte boundary. The read of the APPLICATION's store is
// rounded OUT to whole words and clamped to the store's end, so the widened window can
// never touch a byte the application does not own; it lands in a transient staging
// slice, and only the staging -> dst shift is sub-word - inside our own arena, where an
// unaligned region cannot alias anything else. vkCmdCopyBuffer places no alignment rule
// on a region's offsets or size (unlike vkCmdUpdateBuffer / vkCmdFillBuffer), so the
// shift is a plain legal copy. `dstSkip` is the byte inside `dst` the window starts at.
// This retires `uniform-buffer-byte-tail`.
Bool CopyWireBufferSubWordRangeToSlice(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size,
Uint32 frameIndex, const BufferSlice& dst, Uint64 dstSkip);
void MarkWireBufferGpuWritten(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size);
// P8-D: whether a shader write was marked on this store since the last indirect draw
// read it natively, clearing the mark. The caller records the INDIRECT_COMMAND_READ
// barrier before the draw when this answers true. Transfer writes need no mark: their own
// after-barriers already make them visible to MEMORY_READ at ALL_COMMANDS.
Bool TakeWireIndirectReadBarrier(MG_Pipe::MGPipeHandle res);
// P8-D: every WaitForWireBufferHostAccess that actually waited, over this manager's life -
// the attribution base a caller reads before and after its own ReadWireBuffer.
Uint64 GetWireHostWaitCount() const { return m_wireHostWaits; }
// Resource-op entry points. All run on the server apply owner.
void CreateWireBuffer(MG_Pipe::MGPipeHandle res, const MG_Pipe::MGPResourceDesc& desc);
@@ -123,6 +142,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void FlushWireBuffer(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size, const void* bytes);
void ReadbackWireBuffer(MG_Pipe::MGPipeHandle res, Uint64 offset, Uint64 size);
void DestroyWireBuffer(MG_Pipe::MGPipeHandle res);
// P11 B2 (T0): make the client's AHardwareBuffer `ahb` (a BLOB of `size` bytes) buffer
// `res`'s store - a VkBuffer on memory imported through
// VK_ANDROID_external_memory_android_hardware_buffer, host-visible and mapped like every
// wire store, so every wire path (draw binds, WriteWireBuffer's ordered copy, CPU readers)
// applies unchanged. The T2 store it replaces goes through DeferWireRelease. False = not
// imported (the T2 store is untouched and the client is DECLINED).
Bool ImportWireBuffer(MG_Pipe::MGPipeHandle res, void* ahb, Uint64 size);
// The POST self-test of the sustained-lock pattern on this device (see the definition).
static Bool SelfTestWireImport(char* why, Uint64 whyBytes);
#endif
// Recreate all per-frame transient arenas
@@ -197,6 +225,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// back, and none was persistently mapped, in between - so a memo of resolved
// slices needs no per-buffer re-check. See AcquirePersistentMap for the mapping half.
Uint64 GetSliceEpochCounter() const { return m_sliceEpochCounter; }
#if MOBILEGL_BUILD_DISAGGREGATED
// Bumped every time this manager destroys a WIRE store's VkBuffer (see
// m_wireStoreDestroyEpoch). Unchanged since a memo was taken means no VkBuffer handle
// that memo names can have been freed and re-minted in between, which is the one
// fact a handle-keyed memo of wire descriptors needs and cannot read off the handle.
Uint64 GetWireStoreDestroyEpoch() const { return m_wireStoreDestroyEpoch; }
#endif
// Highest frame serial whose GPU work is known complete; serials at or
// below it may be considered signaled. Drives IsResourceBusy and the
// backend GL fence objects.
@@ -217,16 +252,117 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkBufferObject buffer;
Uint64 size = 0;
Uint64 lastUseSerial = 0;
// P7 wave 2 package B3: the submission expected to carry the most recent GPU use
// of this buffer - the busy predicate's DEFENCE term, not its guarantee (see
// WriteWireBuffer). IsSubmitIndexComplete polls the real fence and reports an
// unsubmitted index as incomplete, but the stamp is taken before the draw is
// recorded and a mid-draw flush can submit it without the draw.
Uint64 lastUseSubmitIndex = 0;
Bool gpuWritesPending = false;
// Only ranges actually submitted by resource_subdata are covered. No
// shadow is retained: flush cannot replay stale bytes over GPU writes.
Vector<Range1D> stagedCoverage;
// P11 B2 (T0): `buffer` is the client's imported AHardwareBuffer. Its readback posts
// no bytes (the client reads its own pages once the wait below it is done).
Bool imported = false;
// P8-D: a shader write was marked (MarkWireBufferGpuWritten) and no indirect draw has
// recorded its INDIRECT_COMMAND_READ barrier since (TakeWireIndirectReadBarrier).
Bool indirectReadBarrierPending = false;
};
// P8-D: see GetWireHostWaitCount.
Uint64 m_wireHostWaits = 0;
static Uint64 WireBufferKey(MG_Pipe::MGPipeHandle res) {
return (static_cast<Uint64>(res.Gen) << 32) | res.Slot;
}
WireBufferResource* FindWireBuffer(MG_Pipe::MGPipeHandle res);
Bool WaitForWireBufferHostAccess(WireBufferResource& resource);
// P11 B2: the serial a T0 store's release waits behind (never "destroy at once").
Uint64 T0ReleaseSerial(Uint64 lastUseSerial) const;
// P7 wave 4 (M2), ID-P7-27: THE WIRE ARM'S ORPHANS NEED A RECLAIM THAT IS NOT A FRAME
// BOUNDARY.
//
// RespecifyWireBuffer - which is every glBufferData that crosses the wire - orphans the
// old store unconditionally, because unlike OnRespecify it has no shadow to upload in
// place from and no cheap way to know the client is re-sending the same size. The orphan
// is legitimate (M1 measured the conditional-orphan mirror: 6701 live against 6702).
// What was missing is the RECLAIM. DeferRelease parks into m_deferredBufferReleases,
// whose only sweep is CollectDeferredReleases from a frame boundary, and on a pbuffer
// replay the server sees ONE present record for the whole run - so on
// minecraft-1.21.4-fabric-iris-bsl-esc-menu-854 the buckets held 25,923 dead stores
// against 28 live wire buffers, one memfd mapping each, and the server died in scudo's
// secondary allocator.
//
// So a wire store is parked HERE instead, with the facts that say when it is dead, and
// THE DEFER PATH ITSELF RECLAIMS - every park sweeps, and the frame boundary is only one
// more sweep point (CollectDeferredReleases), never the one this depends on:
//
// * `lastUseSerial == 0` - read BEFORE RespecifyWireBuffer zeroes it. Zero means no
// GPU command has named the store since it was minted or since the last host-access
// wait proved every recorded command complete (WaitForWireBufferHostAccess); every
// path that hands the store to the GPU stamps m_frameSerial, which starts at 1. Such
// a store is destroyed at the park and never enters the list.
// * `submitIndex` - the renderer's GetSyncPointSubmitIndex() at park time, which is by
// construction the LAST submission that can name this store: every command that
// names it was recorded before the park (the record left behind a bumped slice
// epoch, so no memo can hand it out again), and a recorded command is either already
// submitted (<= m_submitCounter) or in the batch that becomes m_submitCounter + 1.
// IsSubmitIndexComplete(submitIndex) is a fence observation, so this is the gate
// that empties the set MID-FRAME - the shape CollectWireObjects already uses for
// the wire image/view tables.
// * THE RENDERER IS IDLE - IsSubmitIndexComplete(GetSyncPointSubmitIndex()): nothing
// is recorded and every submission has retired. Then every parked store is dead,
// including one tagged for a batch that was abandoned rather than submitted (a
// minimized Present drops its recording), whose own index may never complete.
//
// THE FRAME-SERIAL FLOOR IS DELIBERATELY NOT A PROOF HERE. It cannot move inside a
// frame (NotifyFrameSerialComplete refuses the current serial), so it frees nothing in
// the one-present replay this exists for; the submit index above is the fact that CAN
// move mid-frame, and it is a fence observation rather than a count.
//
// AND EVERY DESTROY HERE HAPPENS MID-FRAME, which the memos above this manager were not
// written for. UniformManager's descriptor memos are keyed on the VkBuffer HANDLE, and
// before M2 a wire store only ever died at the same boundary that clears those memos
// (UniformManager::BeginFrame, from Present or a drain). A store destroyed here can
// have its handle value re-minted by the next Create - a heap pointer under lavapipe -
// while a memo still maps that handle to a descriptor set baked to the dead store's
// memory: silent wrong bytes, no Fatal (ID-P7-43). m_wireStoreDestroyEpoch is the
// fact the memos fold in: every wire-store destroy bumps it, so a memo taken before
// the destroy cannot match after it. Deliberately NOT m_sliceEpochCounter, which every
// WriteWireBuffer bumps and which would defeat the memo on every glBufferSubData.
struct DeferredWireRelease {
VkBufferObject buffer;
Uint64 lastUseSerial = 0;
Uint64 submitIndex = 0;
// Cached: GetSize() is gone once the object is destroyed, and the watermark's
// running total has to stay exact as entries leave.
Uint64 bytes = 0;
};
// Park a wire store, sweep, then hold the watermark. `lastUseSerial` is the releasing
// resource's, captured before the caller resets it.
void DeferWireRelease(VkBufferObject&& buffer, Uint64 lastUseSerial);
// Destroy every parked store proven dead by the rules above. Returns how many.
SizeT SweepDeferredWireReleases();
// MOBILEGL_IPC_WIRE_DEFERRED_MB (Config.h has the semantics): when the parked bytes
// still exceed the budget after a sweep, take the sync point WaitForWireBufferHostAccess
// takes - flush what is recorded, wait for it - and sweep again, which retires every
// parked store, because none can be tagged past the sync point it just waited out.
void EnforceWireDeferredWatermark();
// ...and the same sync point when more than this many stores are parked, whatever their
// bytes (see the definition for the measurement). Not a knob: it bounds an object count
// the byte budget cannot see, and MOBILEGL_IPC_WIRE_DEFERRED_MB=0 disables it too.
static constexpr SizeT kWireDeferredCountCeiling = 1024;
// Teardown: the caller has proven the device idle (Shutdown / RecreateTransientArenas).
void DestroyAllDeferredWireReleases();
// THE one Fatal{ResourceUnavailable, "buffer-write-sync"} site (rule I: no second abort
// for the census to count), shared by the host write that cannot wait for the GPU and
// the watermark that cannot. `site` says which.
[[noreturn]] static void WireBufferSyncFatal(const char* site);
// The wbuf[] gauges (PipeStats.h, Gauge::WireBuffers..WireDeferredSyncs). The peaks are
// taken at the two points the numbers can rise - a park and a mint - and published when
// the stats channel is on; MagmaWireReclaimScenario reads them off the server's line.
void NoteWireStorePeaks();
void PublishWireReclaimGauges();
#endif
Bool InitializeTransientArenas();
static VkBufferUsageFlags GetVkBufferUsage(BufferKind kind);
@@ -263,6 +399,23 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Vector<WeakPtr<VkBufferResource>> m_liveResources;
#if MOBILEGL_BUILD_DISAGGREGATED
std::unordered_map<Uint64, WireBufferResource> m_wireBuffers;
// See DeferredWireRelease. ONE FLAT LIST rather than the per-frame-slot buckets above:
// the whole point is that these entries do not wait for a frame slot to come round.
Vector<DeferredWireRelease> m_deferredWireReleases;
// Bytes currently parked in that list, kept exact so the watermark needs no walk.
Uint64 m_deferredWireBytes = 0;
// Live VkBuffers this arm owns: the stores held by m_wireBuffers plus the parked ones.
// Maintained rather than counted, because the publish runs on every park.
Uint64 m_wireStoreCount = 0;
// Run maxima of the two numbers above and the watermark's sync count, for the gauges.
Uint64 m_wireStoreCountPeak = 0;
Uint64 m_deferredWireBytesPeak = 0;
Uint64 m_wireDeferredSyncs = 0;
// See GetWireStoreDestroyEpoch and the DeferredWireRelease comment. Bumped on every
// path that destroys a wire store's VkBuffer, and never reset (not even by Shutdown),
// for m_sliceEpochCounter's reason: a memo taken before a re-initialize must not match
// a handle minted after it.
Uint64 m_wireStoreDestroyEpoch = 0;
#endif
// Size m_liveResources had just after the last sweep; the next sweep waits for it to double.
SizeT m_liveResourcesLastPruned = 0;
@@ -8,13 +8,45 @@
#include "VkBufferObject.h"
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Remote/Transport/AdoptT0.h>
#endif
namespace MobileGL::MG_Backend::DirectVulkan {
#if MOBILEGL_BUILD_DISAGGREGATED
void VkBufferObject::MoveExternalFrom(VkBufferObject& other) {
m_device = other.m_device;
m_externalMemory = other.m_externalMemory;
m_externalAhb = other.m_externalAhb;
m_externalCoherent = other.m_externalCoherent;
other.m_device = VK_NULL_HANDLE;
other.m_externalMemory = VK_NULL_HANDLE;
other.m_externalAhb = nullptr;
other.m_externalCoherent = true;
}
void VkBufferObject::AdoptExternal(VkDevice device, VkBuffer buffer, VkDeviceMemory memory, VkDeviceSize size,
void* mapped, Bool coherent, void* ahb) {
Destroy();
m_device = device;
m_buffer = buffer;
m_externalMemory = memory;
m_mappedData = mapped;
m_size = size;
m_externalCoherent = coherent;
m_externalAhb = ahb;
}
#endif
VkBufferObject::VkBufferObject(VkBufferObject&& other) noexcept {
m_allocator = other.m_allocator;
m_buffer = other.m_buffer;
m_allocation = other.m_allocation;
m_mappedData = other.m_mappedData;
m_size = other.m_size;
#if MOBILEGL_BUILD_DISAGGREGATED
MoveExternalFrom(other);
#endif
other.m_allocator = nullptr;
other.m_buffer = VK_NULL_HANDLE;
@@ -35,6 +67,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
m_allocation = other.m_allocation;
m_mappedData = other.m_mappedData;
m_size = other.m_size;
#if MOBILEGL_BUILD_DISAGGREGATED
MoveExternalFrom(other);
#endif
other.m_allocator = nullptr;
other.m_buffer = VK_NULL_HANDLE;
@@ -89,6 +124,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
void VkBufferObject::Destroy() {
#if MOBILEGL_BUILD_DISAGGREGATED
if (m_externalMemory != VK_NULL_HANDLE) {
// P11 B2 (T0): an imported AHardwareBuffer store. Reached through the wire arm's
// deferred release, i.e. once no submission can still name it.
if (m_mappedData != nullptr) vkUnmapMemory(m_device, m_externalMemory);
if (m_buffer != VK_NULL_HANDLE) vkDestroyBuffer(m_device, m_buffer, nullptr);
vkFreeMemory(m_device, m_externalMemory, nullptr);
MG_Remote::Transport::AdoptT0::ReleaseImported(m_externalAhb);
m_mappedData = nullptr;
m_buffer = VK_NULL_HANDLE;
m_externalMemory = VK_NULL_HANDLE;
m_externalAhb = nullptr;
m_device = VK_NULL_HANDLE;
m_externalCoherent = true;
m_size = 0;
return;
}
#endif
Unmap();
if (m_allocator != nullptr && m_buffer != VK_NULL_HANDLE && m_allocation != nullptr) {
vmaDestroyBuffer(m_allocator, m_buffer, m_allocation);
@@ -117,6 +170,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
void VkBufferObject::Unmap() {
#if MOBILEGL_BUILD_DISAGGREGATED
// An imported store stays mapped for its life (Destroy unmaps it).
if (m_externalMemory != VK_NULL_HANDLE) return;
#endif
if (!IsValid() || m_mappedData == nullptr) {
m_mappedData = nullptr;
return;
@@ -143,6 +200,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
Memcpy(static_cast<Uint8*>(mapped) + offset, data, static_cast<SizeT>(size));
#if MOBILEGL_BUILD_DISAGGREGATED
if (m_externalMemory != VK_NULL_HANDLE) {
if (m_externalCoherent) return true;
VkMappedMemoryRange range{VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE};
range.memory = m_externalMemory;
range.offset = 0;
range.size = VK_WHOLE_SIZE;
return vkFlushMappedMemoryRanges(m_device, 1, &range) == VK_SUCCESS;
}
#endif
const VkResult flushResult = vmaFlushAllocation(m_allocator, m_allocation, offset, size);
if (flushResult != VK_SUCCESS) {
MGLOG_E_ONCE("VkBufferObject::Upload failed: vmaFlushAllocation returned %d", flushResult);
@@ -167,6 +234,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (resolvedSize == 0) {
return true;
}
#if MOBILEGL_BUILD_DISAGGREGATED
if (m_externalMemory != VK_NULL_HANDLE) {
if (m_externalCoherent) return true;
VkMappedMemoryRange range{VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE};
range.memory = m_externalMemory;
range.offset = 0;
range.size = VK_WHOLE_SIZE;
return vkInvalidateMappedMemoryRanges(m_device, 1, &range) == VK_SUCCESS;
}
#endif
const VkResult result = vmaInvalidateAllocation(m_allocator, m_allocation, offset, resolvedSize);
if (result != VK_SUCCESS) {
@@ -64,7 +64,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
void* GetMappedData() const { return m_mappedData; }
Bool IsMapped() const { return m_mappedData != nullptr; }
Bool IsValid() const { return m_allocator != nullptr && m_buffer != VK_NULL_HANDLE && m_allocation != nullptr; }
Bool IsValid() const {
#if MOBILEGL_BUILD_DISAGGREGATED
if (m_externalMemory != VK_NULL_HANDLE) return m_buffer != VK_NULL_HANDLE;
#endif
return m_allocator != nullptr && m_buffer != VK_NULL_HANDLE && m_allocation != nullptr;
}
#if MOBILEGL_BUILD_DISAGGREGATED
// P11 B2 (T0): a store whose memory is NOT this allocator's - a client's AHardwareBuffer
// imported with VK_ANDROID_external_memory_android_hardware_buffer into `memory` (a
// dedicated allocation bound to `buffer`, mapped persistently at `mapped`). It behaves
// like any mapped store (GetSlice, Upload, Invalidate), and Destroy frees the buffer and
// the memory and drops `ahb` - which is why the store goes through the wire arm's
// serial-gated deferred release like every other: the AHardwareBuffer is let go only once
// no submission can still name it. `coherent` = the memory type is HOST_COHERENT.
void AdoptExternal(VkDevice device, VkBuffer buffer, VkDeviceMemory memory, VkDeviceSize size,
void* mapped, Bool coherent, void* ahb);
Bool IsExternal() const { return m_externalMemory != VK_NULL_HANDLE; }
#endif
private:
VmaAllocator m_allocator = nullptr;
@@ -72,5 +90,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VmaAllocation m_allocation = nullptr;
void* m_mappedData = nullptr;
VkDeviceSize m_size = 0;
#if MOBILEGL_BUILD_DISAGGREGATED
VkDevice m_device = VK_NULL_HANDLE;
VkDeviceMemory m_externalMemory = VK_NULL_HANDLE;
void* m_externalAhb = nullptr;
Bool m_externalCoherent = true;
void MoveExternalFrom(VkBufferObject& other);
#endif
};
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -23,6 +23,8 @@
#include <MG_Pipe/PipeApply.h>
#include <MG_Remote/Server/StagedTextureStore.h>
#include "../DirectVulkan.h"
// P7 wave 2 package B3: rule I's tally for the silent exits on this file's wire arm.
#include "WireDeclineTally.h"
#endif
#include <algorithm>
#include <cstdio>
@@ -1331,6 +1333,42 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
VkTextureManager::TextureResource* VkTextureManager::FindTextureResource(
MG_State::GLState::ITextureObject& textureOrView) {
auto it = m_textureResources.find(MakeTextureIdentity(&StorageTextureOf(textureOrView)));
return it != m_textureResources.end() ? &it->second : nullptr;
}
Bool VkTextureManager::HasLiveImage(MG_State::GLState::ITextureObject& textureOrView) const {
auto it = m_textureResources.find(MakeTextureIdentity(&StorageTextureOf(textureOrView)));
return it != m_textureResources.end() && it->second.image != VK_NULL_HANDLE;
}
Bool VkTextureManager::HasPendingTexelUpload(MG_State::GLState::ITextureObject& textureOrView) const {
auto& storage = StorageTextureOf(textureOrView);
auto it = m_textureResources.find(MakeTextureIdentity(&storage));
if (it == m_textureResources.end() || it->second.image == VK_NULL_HANDLE) {
return false;
}
return it->second.syncedShapeVersion == storage.GetShapeVersion() &&
it->second.syncedContentVersion != storage.GetContentVersion();
}
Bool VkTextureManager::WasTextureWrittenThisRecording(MG_State::GLState::ITextureObject& textureOrView) const {
auto it = m_textureResources.find(MakeTextureIdentity(&StorageTextureOf(textureOrView)));
return it != m_textureResources.end() && it->second.lastRecordingWriteGeneration == m_recordingGeneration;
}
void VkTextureManager::StampTextureRecordingWrite(MG_State::GLState::ITextureObject* texture) {
if (texture == nullptr) {
return;
}
auto it = m_textureResources.find(MakeTextureIdentity(&StorageTextureOf(*texture)));
if (it != m_textureResources.end()) {
StampResourceRecordingWrite(it->second);
}
}
void VkTextureManager::UpdateTrackedImageLayout(MG_State::GLState::ITextureObject* texture, VkImageLayout newLayout) {
MOBILEGL_ASSERT(texture != nullptr, "UpdateTrackedImageLayout: texture is null");
auto it = m_textureResources.find(MakeTextureIdentity(texture));
@@ -1359,7 +1397,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
"UpdateTrackedImageLayoutAfterAttachmentWrite: textureId=%d mipLevel=%u out of range %u",
texture->GetExternalIndex(), writtenMipLevel, resource.mipLevels);
// Pre-pass stream bookkeeping: the render pass that just ended wrote this image.
StampResourceRecordingUse(resource);
StampResourceRecordingWrite(resource);
if (resource.layout != newLayout && resource.mipLevels > 1) {
VkPipelineStageFlags srcStageMask = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
@@ -1473,8 +1511,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
resource->aspect, 0, resource->mipLevels);
MOBILEGL_ASSERT(ok, "TransitionTextureForStorageImage: transition failed for textureId=%d",
texture.GetExternalIndex());
// Pre-pass stream bookkeeping: a command referencing the image was recorded.
StampResourceRecordingUse(*resource);
// Pre-pass stream bookkeeping: a command referencing the image was recorded, and an
// image unit may store to it.
StampResourceRecordingWrite(*resource);
return ok;
}
@@ -1805,8 +1844,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// SyncTexture reads the frontend ITextureObject: the target, the format, the level
// chain, the dirty scan and - through UploadDirtyMipLevels - the CLIENT's mip shadow
// texels. Under an active transport none of that memory is the server's to read
// (rule E), which is exactly what Magma's texture-legacy-arm scope has been waiving
// since P5c. This arm is the same sync re-sourced: the SHAPE comes from the applier's
// (rule E), which is exactly what Magma's texture-legacy-arm scope waived from P5c until
// P7 retired it (SyncTexture). This arm is the same sync re-sourced: the SHAPE comes from the applier's
// resource record (MGPipeApplier().TextureResources - the resource_create /
// resource_respecify descriptors), the TEXELS from the server's staged-texture store
// (P5c tx, adopted at apply time), and the WHICH-LEVELS from the record's
@@ -2069,13 +2108,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (desc.StorageKind != static_cast<Uint8>(TextureStorageType::Mipmap)) {
MGLOG_W_ONCE("Magma wire texture {slot=%u, gen=%u}: buffer-backed textures are not migrated; declined",
desc.Resource.Slot, desc.Resource.Gen);
WireDeclineTally::Count(WireDeclineSite::ShapeBufferBacked);
return false;
}
// ResolveWireTextureStorage has already followed ViewOf. Views share the storage
// owner's image and layout; allocating a second image would break aliasing.
if (!MG_Pipe::MGPipeHandleIsNull(desc.ViewOf)) return false;
if (!MG_Pipe::MGPipeHandleIsNull(desc.ViewOf)) {
MGL_WIRE_DECLINE_AT(ShapeViewOfStorage,
"texture {slot=%u, gen=%u} is a view whose storage owner did not resolve",
desc.Resource.Slot, desc.Resource.Gen);
return false;
}
if (desc.Levels == 0) {
// A create before the first respecify: no storage, nothing to back.
MGL_WIRE_DECLINE_AT(ShapeNoLevels, "texture {slot=%u, gen=%u} has no defined level yet",
desc.Resource.Slot, desc.Resource.Gen);
return false;
}
TextureShapeInfo shapeInfo{};
@@ -2084,6 +2131,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
"(%ux%ux%u, %u layers); declined",
desc.Resource.Slot, desc.Resource.Gen, static_cast<Uint32>(desc.Target), desc.Width,
desc.Height, desc.Depth, static_cast<Uint32>(desc.ArrayLayers));
WireDeclineTally::Count(WireDeclineSite::ShapeNoVulkanShape);
return false;
}
const TextureInternalFormat internalFormat = static_cast<TextureInternalFormat>(desc.InternalFormat);
@@ -2092,6 +2140,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (format == VK_FORMAT_UNDEFINED) {
MGLOG_W_ONCE("Magma wire texture {slot=%u, gen=%u}: no backing VkFormat for internal format 0x%x",
desc.Resource.Slot, desc.Resource.Gen, desc.InternalFormat);
WireDeclineTally::Count(WireDeclineSite::ShapeNoVkFormat);
return false;
}
// X8_D24 lacks optimal-tiling support on several drivers (lavapipe included); the same
@@ -2114,6 +2163,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (!TryResolveSampleCountFlagBits(static_cast<Int>(desc.Samples), resolvedSampleCount)) {
MGLOG_W_ONCE("Magma wire texture {slot=%u, gen=%u}: unsupported sample count %u",
desc.Resource.Slot, desc.Resource.Gen, static_cast<Uint32>(desc.Samples));
WireDeclineTally::Count(WireDeclineSite::ShapeSampleCount);
return false;
}
// glTexStorage*Multisample(samples = 1) is legal GL, but a one-sample image cannot
@@ -2249,7 +2299,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// old image must reach the queue first; waiting only for the preservation
// fence cannot order work that is still in the renderer's open recording.
// The descriptor caller preflights this before capturing its command buffer.
if (preserve && pVulkanRenderer && !pVulkanRenderer->FlushWirePendingCommandsForTextureUpdate()) return false;
if (preserve && pVulkanRenderer && !pVulkanRenderer->FlushWirePendingCommandsForTextureUpdate()) {
MGL_WIRE_DECLINE_AT(ShapePreserveFlushFailed,
"texture {slot=%u, gen=%u}: earlier work could not be submitted before "
"preserving its texels across a reallocation",
desc.Resource.Slot, desc.Resource.Gen);
return false;
}
TextureResource replacement;
VkImageCreateInfo imageInfo{};
@@ -2296,9 +2352,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
if (imageFormatResult != VK_SUCCESS ||
(isMultisample && (imageFormatProperties.sampleCounts & resolvedSampleCount) == 0)) {
MGLOG_D("%s: image flags=0x%x sampleCount=%d are unsupported for texture {slot=%u, gen=%u}",
__func__, static_cast<Uint32>(imageInfo.flags),
static_cast<Int>(imageInfo.samples), desc.Resource.Slot, desc.Resource.Gen);
// Was an MGLOG_D, which a Release build compiles away, so on the builds that ship
// this exit left no line - one of five Shape* exits that did (with ViewOf,
// NoLevels, preserve-flush and preserve-copy) before B3's review round.
MGL_WIRE_DECLINE_AT(ShapeImageFlagsUnsupported,
"texture {slot=%u, gen=%u}: image flags=0x%x sampleCount=%d are unsupported "
"by the device",
desc.Resource.Slot, desc.Resource.Gen, static_cast<Uint32>(imageInfo.flags),
static_cast<Int>(imageInfo.samples));
return false;
}
}
@@ -2317,6 +2378,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
imageInfo.extent.height, imageInfo.extent.depth, imageInfo.arrayLayers,
imageInfo.mipLevels, static_cast<Int>(imageInfo.samples),
static_cast<Int>(imageInfo.format));
WireDeclineTally::Count(WireDeclineSite::ShapeCreateImageFailed);
return false;
}
++m_textureImageEpoch; // a new attachment image invalidates cached render passes
@@ -2338,7 +2400,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (preserve) {
FlushPendingUploads();
if (!PreserveTextureContentsOnRecreate(m_device, m_commandPool, m_graphicsQueue,
resource, replacement, true)) return false;
resource, replacement, true)) {
MGL_WIRE_DECLINE_AT(ShapePreserveCopyFailed,
"texture {slot=%u, gen=%u}: its texels could not be copied into the "
"reallocated image",
desc.Resource.Slot, desc.Resource.Gen);
return false;
}
}
DeferResourceRelease(Move(resource));
std::destroy_at(&resource);
@@ -2385,6 +2453,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MGLOG_E_ONCE("Magma wire texture {slot=%u, gen=%u}: pending upload for level %u exceeds the "
"image's %u levels; left pending",
handle.Slot, handle.Gen, static_cast<Uint32>(level), resource.mipLevels);
WireDeclineTally::Count(WireDeclineSite::UploadLevelBeyondChain);
continue;
}
const IntVec3 glExtent = store.LevelExtentOrUndefined(key, uploadTarget, level);
@@ -2405,9 +2474,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
texelSize.y() != static_cast<Int>(std::max(resource.extent.height >> level, 1u)) ||
(volume && texelSize.z() != static_cast<Int>(std::max(resource.depth >> level, 1u))) ||
(!volume && Uint64(ResolveUploadArrayLayer(static_cast<TextureUploadTarget>(uploadTarget))) +
static_cast<Uint64>(texelSize.z()) > resource.arrayLayers))
static_cast<Uint64>(texelSize.z()) > resource.arrayLayers)) {
// NOT `incomplete`: this arm reports SUCCESS while leaving the entry pending,
// so the draw proceeds against texels that were never uploaded. Counted
// separately for exactly that reason (B3).
MGL_WIRE_DECLINE_AT(UploadLevelExtentMismatch,
"texture {slot=%u, gen=%u} level %u: the staged extent does not match the "
"image's extent for that level; left pending and NOT uploaded",
handle.Slot, handle.Gen, static_cast<Uint32>(level));
continue;
}
if (texelSize.x() <= 0 || texelSize.y() <= 0 || byteSize == 0) {
MGL_WIRE_DECLINE_AT(UploadEmptyLevel, "texture {slot=%u, gen=%u} level %u is empty",
handle.Slot, handle.Gen, static_cast<Uint32>(level));
incomplete = true;
continue;
}
@@ -2423,6 +2502,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
"level %u left pending",
handle.Slot, handle.Gen, static_cast<Int>(resource.format),
static_cast<Uint32>(level));
WireDeclineTally::Count(WireDeclineSite::UploadConversion);
incomplete = true;
continue;
}
@@ -2435,6 +2515,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
(resource.aspect & VK_IMAGE_ASPECT_STENCIL_BIT);
if (vkuFormatIsCompressed(resource.format) || !levelTexels ||
item.byteSize % levelTexels != 0 || item.byteSize / levelTexels == 0) {
MGL_WIRE_DECLINE_AT(UploadTexelArithmetic,
"texture {slot=%u, gen=%u} level %u: compressed, or its byte size is not a "
"whole number of texels",
handle.Slot, handle.Gen, static_cast<Uint32>(level));
incomplete = true;
continue;
}
@@ -2454,6 +2538,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (offset[axis] < 0 || extent[axis] <= 0 ||
static_cast<Uint64>(offset[axis]) + static_cast<Uint64>(extent[axis]) >
static_cast<Uint64>(texelSize[axis])) {
MGL_WIRE_DECLINE_AT(UploadRegionOutOfBounds,
"texture {slot=%u, gen=%u} level %u: a staged region lies outside "
"the level",
handle.Slot, handle.Gen, static_cast<Uint32>(level));
incomplete = true;
return;
}
@@ -2496,7 +2584,25 @@ namespace MobileGL::MG_Backend::DirectVulkan {
addRegion({region.X, region.Y, region.Z, region.W, region.H, region.D});
}
}
if (incomplete) return false;
// B3: the pending gauge is sampled here, on every pass through the funnel, because a
// refusal below does NOT consume the entries - they and their level shadows stay
// alive for the next draw to trip over (and, on a heavy-texture trace, to accumulate).
{
Uint64 pendingBytes = 0;
for (const auto& pending : record.PendingUploads) {
pendingBytes += static_cast<Uint64>(
store.LevelByteSize(MG_Remote::Server::StagedTextureStore::KeyForHandle(handle),
pending.UploadTarget, pending.Level));
}
WireDeclineTally::SetPendingGauge(static_cast<Uint64>(record.PendingUploads.size()), pendingBytes);
}
if (incomplete) {
MGL_WIRE_DECLINE_AT(UploadIncomplete,
"texture {slot=%u, gen=%u}: at least one pending level could not be staged; "
"every draw that samples it is declined until it can",
handle.Slot, handle.Gen);
return false;
}
if (items.empty()) {
return true;
}
@@ -2505,7 +2611,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Submit older renderer work before adding the new bytes to the batch,
// otherwise Draw(old T), TexSubImage(T), Draw(new T) can upload before
// the first draw. A clean texture never reaches this submission boundary.
if (pVulkanRenderer && !pVulkanRenderer->FlushWirePendingCommandsForTextureUpdate()) return false;
if (pVulkanRenderer && !pVulkanRenderer->FlushWirePendingCommandsForTextureUpdate()) {
MGL_WIRE_DECLINE_AT(UploadFlushFailed,
"texture {slot=%u, gen=%u}: earlier renderer work could not be submitted before "
"its upload batch",
handle.Slot, handle.Gen);
return false;
}
// UploadDirtyMipLevels' batching rules, verbatim: flush an open batch that already
// writes this image and was touched by the open recording, and bound the staging bytes
@@ -2602,9 +2714,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MG_Pipe::MGPipeHandle VkTextureManager::ResolveWireTextureStorage(
MG_Pipe::MGPipeHandle handle, Uint32& level, Uint32& layer,
VkFormat* viewFormat, Uint32* layerCount) {
VkFormat* viewFormat, Uint32* layerCount, Bool* outsideWindow) {
const auto& state = MG_Pipe::MGPipeApplier();
if (viewFormat) *viewFormat = VK_FORMAT_UNDEFINED;
if (outsideWindow) *outsideWindow = false;
const auto outside = [outsideWindow] {
if (outsideWindow) *outsideWindow = true;
return MG_Pipe::kMGPipeNullHandle;
};
Uint32 mappedLevel = level, mappedLayer = layer;
Uint32 remainingLayers = ~Uint32{0};
VkFormat outerFormat = VK_FORMAT_UNDEFINED;
@@ -2616,12 +2733,13 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const auto& record = state.TextureResources[handle.Slot];
if (!record.Live || record.Gen != handle.Gen) return MG_Pipe::kMGPipeNullHandle;
if (MG_Pipe::MGPipeHandleIsNull(record.Desc.ViewOf)) {
if (mappedLevel >= record.Desc.Levels) return MG_Pipe::kMGPipeNullHandle;
// Levels == 0 is a texture with no storage at all: every level is outside it.
if (mappedLevel >= record.Desc.Levels) return outside();
const Uint32 layers = record.Desc.Target ==
static_cast<Uint8>(MG_Pipe::MGPipeResourceTarget::Tex3D)
? std::max(record.Desc.Depth >> mappedLevel, 1u)
: std::max<Uint32>(record.Desc.ArrayLayers, 1u);
if (mappedLayer >= layers) return MG_Pipe::kMGPipeNullHandle;
if (mappedLayer >= layers) return outside();
level = mappedLevel;
layer = mappedLayer;
if (viewFormat) *viewFormat = outerFormat;
@@ -2633,9 +2751,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return MG_Pipe::kMGPipeNullHandle;
const auto& viewRecord = state.SamplerViewCsos[viewHandle.Slot];
const auto& view = viewRecord.View;
if (!viewRecord.Live || viewRecord.Gen != viewHandle.Gen || view.Texture != handle ||
mappedLevel >= view.NumLevels || mappedLayer >= view.NumLayers)
if (!viewRecord.Live || viewRecord.Gen != viewHandle.Gen || view.Texture != handle)
return MG_Pipe::kMGPipeNullHandle;
if (mappedLevel >= view.NumLevels || mappedLayer >= view.NumLayers) return outside();
if (outerFormat == VK_FORMAT_UNDEFINED) {
outerFormat = ResolveTextureFormatInfo(static_cast<TextureInternalFormat>(view.InternalFormat)).format;
if (outerFormat == VK_FORMAT_UNDEFINED) return MG_Pipe::kMGPipeNullHandle;
@@ -2659,7 +2777,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint32 level = 0, layer = 0;
VkFormat viewFormat = VK_FORMAT_UNDEFINED;
const auto storage = ResolveWireTextureStorage(handle, level, layer, &viewFormat);
if (MG_Pipe::MGPipeHandleIsNull(storage)) return nullptr;
if (MG_Pipe::MGPipeHandleIsNull(storage)) {
MGL_WIRE_DECLINE_AT(TexResolveStorage,
"texture {slot=%u, gen=%u}: no storage owner resolved", handle.Slot,
handle.Gen);
return nullptr;
}
if (storage != handle) {
auto* resource = SyncTextureResourceByHandle(storage, false, requireStorage);
if (!resource) return nullptr;
@@ -2668,6 +2791,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
vkuFormatCompatibilityClass(viewFormat) != vkuFormatCompatibilityClass(resource->format))) {
MGLOG_E_ONCE("Magma wire view: format %d cannot reinterpret storage format %d",
static_cast<Int>(viewFormat), static_cast<Int>(resource->format));
WireDeclineTally::Count(WireDeclineSite::TexViewFormat);
return nullptr;
}
return resource;
@@ -2678,6 +2802,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
if (handle.Slot >= records.size()) {
MGLOG_E_ONCE("Magma wire texture {slot=%u, gen=%u}: no applier record at that slot; declined",
handle.Slot, handle.Gen);
WireDeclineTally::Count(WireDeclineSite::TexNoApplierRecord);
return nullptr;
}
auto& record = records[handle.Slot];
@@ -2687,6 +2812,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MGLOG_E_ONCE("Magma wire texture {slot=%u, gen=%u}: the applier's record is %s; declined",
handle.Slot, handle.Gen,
!record.Live ? "dead" : "of another generation");
WireDeclineTally::Count(WireDeclineSite::TexRecordDeadOrStale);
return nullptr;
}
const Uint64 key = MG_Remote::Server::StagedTextureStore::KeyForHandle(handle);
@@ -2698,12 +2824,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
return &resource;
}
if (!SyncWireTextureShape(record, resource, requireStorage)) {
MGL_WIRE_DECLINE_AT(TexShapeSync, "texture {slot=%u, gen=%u}: no backing image; see the Shape* tally",
handle.Slot, handle.Gen);
return nullptr;
}
if (!renderbuffer && !UploadPendingWireLevels(handle, record, resource)) {
MGL_WIRE_DECLINE_AT(TexUploadPendingLevels,
"texture {slot=%u, gen=%u}: pending levels could not be uploaded; see the "
"Upload* tally",
handle.Slot, handle.Gen);
return nullptr;
}
if (requireStorage && (resource.usageFlags & VK_IMAGE_USAGE_STORAGE_BIT) == 0) {
MGL_WIRE_DECLINE_AT(TexStorageUsage,
"texture {slot=%u, gen=%u} is bound as a shader image but its allocation has no "
"STORAGE usage",
handle.Slot, handle.Gen);
return nullptr;
}
if (requireStorage && (resource.usageFlags & VK_IMAGE_USAGE_STORAGE_BIT) == 0) return nullptr;
resource.syncedWireSerial = record.Serial;
return &resource;
}
@@ -2863,16 +3001,16 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool VkTextureManager::SyncTexture(MG_State::GLState::ITextureObject &texture,
TextureResource &outResource) {
#if MOBILEGL_BUILD_DISAGGREGATED
// P5c (gt): Magma's texture sync still reads - and clears - the CLIENT's mip shadow:
// the dirty scan below, and UploadDirtyMipLevels' texel reads / region reads /
// MarkStorageDirty(false) clears. CONTRACT-P5C §2 migrated Espryt's sync and Magma's
// T5 writes, and left THIS path on the legacy arm; under the verb barrier and one
// address space the reads answer correctly, and the migration is P7's server-side
// sync. The scope is the debt's greppable form (MipmapStorage.h); outside it the
// layer-1 guard still aborts.
const MG_State::GLState::MGPipeTextureLegacyArmScope textureLegacyArm;
#endif
// This is the FRONTEND-object sync: it reads - and clears - the client's mip shadow (the
// dirty scan below, UploadDirtyMipLevels' texel / region reads and MarkStorageDirty(false)
// clears), so it belongs to the monolith arm only. Under an active transport every wire
// verb syncs through the handle-keyed arm (SyncTextureResourceByHandle, and the sampled
// wire descriptors in UniformManager) from the applier's record and the staged-texture
// store. P5c's named exemption for this function (MGPipeTextureLegacyArmScope) was retired
// in P7 (ratchet88) once a probe proved it never armed on the apply thread; a regression
// now meets the layer-1 guard (Fatal{RoleViolation, "texture-legacy-arm"},
// MipmapStorage.cpp) by accessor name. This block is exactly as many lines as the one it
// replaced: __LINE__ reaches this file's pull .text, and G1 compares the bytes.
// Cross-draw fast path: if the resource is already built and neither the texture's
// pixel content (bumped in MarkStorageDirty), its SHAPE (bumped in BumpShapeVersion)
// nor its params changed since the last sync, there is nothing to re-check or
@@ -3976,18 +4114,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkDeviceSize offset = 0;
// Sub-region upload (a small sprite in a big atlas): only the dirty box
// is staged and copied. texelSize keeps the LEVEL extent - the staging
// row copy needs it for the shadow's stride. Plain color formats only;
// the RGB-expand and depth(+stencil) conversion passes rewrite whole
// levels and stay full-size.
// row copy needs it for the stride. Color formats only (an RGB-expanded
// level is cut from its expanded bytes); the depth(+stencil) conversion
// passes rewrite whole levels and stay full-size.
Bool subRegion = false;
IntVec3 regionLo = {0, 0, 0};
IntVec3 regionSize = {0, 0, 0};
SizeT texelBytes = 0;
// Scatter refinement of the single dirty box: when the storage's rect
// list reports the writes' true footprint (~100 sprites whose union box
// spans the whole atlas), each rect is staged tightly and copied with
// its own VkBufferImageCopy in ONE vkCmdCopyBufferToImage. Empty means
// "stage the one box above". Only set while subRegion.
// The writes' exact footprint when it is more than one box (~100 sprites
// whose union box spans the whole atlas): each rect is staged tightly and
// copied with its own VkBufferImageCopy in ONE vkCmdCopyBufferToImage.
// Empty means "stage the one box above". Only set while subRegion.
Vector<MG_State::GLState::MipmapDirtyRegion> rects;
};
@@ -3999,6 +4136,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
targets.push_back(uploadTarget);
}
const TextureFormatInfo formatInfo = ResolveTextureFormatInfo(mipmapTexture.GetFormat());
Vector<MG_State::GLState::MipmapDirtyRegion> dirtyFootprint;
VkDeviceSize stagingSize = 0;
for (const TextureUploadTarget target : targets) {
@@ -4041,34 +4179,48 @@ namespace MobileGL::MG_Backend::DirectVulkan {
uploadItem.source = source;
uploadItem.offset = stagingSize;
uploadItem.uploadByteSize = byteSize;
if (!formatInfo.expandRgbToRgba &&
GetAspectMaskForFormat(outResource.format) == VK_IMAGE_ASPECT_COLOR_BIT) {
// The RGB -> RGBA expansion rewrites the level texel for texel into a tight buffer
// of its own, so it runs first and the dirty region below is taken from whichever
// bytes are staged. Uploading an expanded level WHOLE put the shadow's stale texels
// over whatever the GPU had drawn into the rest of it.
if (formatInfo.expandRgbToRgba) {
const Bool expanded = ExpandRgbSourceToRgba(source, byteSize, texelSize, formatInfo,
uploadItem.expandedData);
MOBILEGL_ASSERT(expanded,
"UploadDirtyMipLevels: failed to expand RGB textureId=%d target=%s level=%u to RGBA staging data",
mipmapTexture.GetExternalIndex(),
MG_Util::ConvertTextureUploadTargetToString(target).c_str(), level);
uploadItem.uploadByteSize = uploadItem.expandedData.size();
}
const SizeT levelBytes = uploadItem.uploadByteSize;
if (GetAspectMaskForFormat(outResource.format) == VK_IMAGE_ASPECT_COLOR_BIT) {
const auto region = mipmapTexture.GetStorageDirtyRegion(target, level);
const SizeT texelCount = static_cast<SizeT>(texelSize.x()) *
static_cast<SizeT>(texelSize.y()) *
static_cast<SizeT>(std::max(texelSize.z(), 1));
if (!region.Empty() && !region.CoversWholeLevel(texelSize) && texelCount > 0 &&
byteSize % texelCount == 0) {
// Exactly the texels the client wrote, never their union box or the storage's
// rect list (which merges touching writes into their bounding box, folds when
// full and is withheld past 3/4 of the box): the image may hold what the GPU
// drew or cleared anywhere else, and the shadow is stale there. Each box is one
// more VkBufferImageCopy in the same vkCmdCopyBufferToImage, which costs next
// to nothing, so there is no cheaper shape worth being wrong for. A union box
// that spans the level with gaps in it is not a whole-level write either.
mipmapTexture.GetStorageDirtyFootprint(target, level, dirtyFootprint);
if (!region.Empty() && (dirtyFootprint.size() >= 2 || !region.CoversWholeLevel(texelSize)) &&
texelCount > 0 && levelBytes % texelCount == 0) {
uploadItem.subRegion = true;
uploadItem.regionLo = region.lo;
uploadItem.regionSize = {region.hi.x() - region.lo.x(), region.hi.y() - region.lo.y(),
region.hi.z() - region.lo.z()};
uploadItem.texelBytes = byteSize / texelCount;
uploadItem.texelBytes = levelBytes / texelCount;
uploadItem.uploadByteSize = static_cast<SizeT>(uploadItem.regionSize.x()) *
static_cast<SizeT>(uploadItem.regionSize.y()) *
static_cast<SizeT>(uploadItem.regionSize.z()) *
uploadItem.texelBytes;
// Scatter refinement: the storage only hands out its rect list
// when the rects' summed area is materially smaller than the
// union box (0 otherwise), so taking it always stages fewer
// bytes than the box - the very amplification this path exists
// to avoid paying twice.
MG_State::GLState::MipmapDirtyRegion
dirtyRects[MG_State::GLState::MipmapStorage::kMaxDirtyRects];
const SizeT dirtyRectCount = mipmapTexture.GetStorageDirtyRects(
target, level, dirtyRects, MG_State::GLState::MipmapStorage::kMaxDirtyRects);
if (dirtyRectCount >= 2) {
uploadItem.rects.assign(dirtyRects, dirtyRects + dirtyRectCount);
// Several boxes: each is staged tightly and copied on its own. One box is
// the union box above.
if (dirtyFootprint.size() >= 2) {
uploadItem.rects.assign(dirtyFootprint.begin(), dirtyFootprint.end());
SizeT rectTexels = 0;
for (const auto& rect : uploadItem.rects) {
rectTexels += rect.TexelCount();
@@ -4094,15 +4246,6 @@ namespace MobileGL::MG_Backend::DirectVulkan {
}
}
}
if (formatInfo.expandRgbToRgba) {
const Bool expanded = ExpandRgbSourceToRgba(source, byteSize, texelSize, formatInfo,
uploadItem.expandedData);
MOBILEGL_ASSERT(expanded,
"UploadDirtyMipLevels: failed to expand RGB textureId=%d target=%s level=%u to RGBA staging data",
mipmapTexture.GetExternalIndex(),
MG_Util::ConvertTextureUploadTargetToString(target).c_str(), level);
uploadItem.uploadByteSize = uploadItem.expandedData.size();
}
uploadItems.push_back(Move(uploadItem));
if (!uploadItems.back().expandedData.empty()) {
uploadItems.back().source = uploadItems.back().expandedData.data();
@@ -4337,6 +4480,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
const Bool depthSelectsArrayLayer = outResource.viewType == VK_IMAGE_VIEW_TYPE_1D_ARRAY ||
outResource.viewType == VK_IMAGE_VIEW_TYPE_2D_ARRAY ||
outResource.viewType == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY;
Vector<VkBufferImageCopy> rectCopies;
for (const auto& item : uploadItems) {
if (!item.rects.empty()) {
// Multi-rect item: one VkBufferImageCopy per rect, all submitted in a
@@ -4344,7 +4488,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// construction, so no two copies write the same texels. Multi-rect
// implies subRegion, which implies a plain color aspect - the combined
// depth-stencil split below can never see one of these.
VkBufferImageCopy rectCopies[MG_State::GLState::MipmapStorage::kMaxDirtyRects];
rectCopies.resize(item.rects.size());
Uint32 rectCopyCount = 0;
VkDeviceSize runningOffset = item.offset;
for (const auto& rect : item.rects) {
@@ -4375,7 +4519,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
runningOffset += static_cast<VkDeviceSize>(rect.TexelCount() * item.texelBytes);
}
vkCmdCopyBufferToImage(commandBuffer, stagingBuffer, outResource.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, rectCopyCount, rectCopies);
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, rectCopyCount, rectCopies.data());
continue;
}
const Uint32 depthOrLayers = item.texelSize.z() > 0 ? static_cast<Uint32>(item.texelSize.z()) : 1u;
@@ -332,6 +332,11 @@ public:
// into the frame's PRE command buffer - which executes strictly before the
// frame's commands - instead of splitting the active render pass.
Uint64 lastRecordingGeneration = 0;
// The same, for the last recorded command that WRITES the image (a render pass it is an
// attachment of, a clear, a blit/copy into it, a mip generation, a storage-image binding).
// An upload into an image written this recording has to be ordered after that write
// (VulkanRenderer::OrderPendingUploadAfterRecording).
Uint64 lastRecordingWriteGeneration = 0;
// Snapshot of ITextureObject::GetContentVersion() at the last successful sync;
// lets SyncTexture skip the whole re-check/re-upload when content is unchanged.
Uint64 syncedContentVersion = 0;
@@ -381,6 +386,7 @@ public:
std::swap(this->storageUsageResolved, that.storageUsageResolved);
std::swap(this->syncedTextureParamsVersion, that.syncedTextureParamsVersion);
std::swap(this->lastRecordingGeneration, that.lastRecordingGeneration);
std::swap(this->lastRecordingWriteGeneration, that.lastRecordingWriteGeneration);
std::swap(this->syncedContentVersion, that.syncedContentVersion);
std::swap(this->syncedMipLevelCount, that.syncedMipLevelCount);
std::swap(this->syncedShapeVersion, that.syncedShapeVersion);
@@ -566,8 +572,13 @@ public:
// Map a texture/view-relative subresource to its live storage owner. layerCount, when
// supplied, receives the number of accessible layers starting at the input layer.
// viewFormat is UNDEFINED for an ordinary texture, or the outer view's format.
// outsideWindow, when supplied, says WHY a null came back: true only when every record on
// the way was live and consistent and the (level, layer) simply is not there - past the
// owner's levels or layers, or past a view's own window. A dead record, a missing or stale
// view CSO and an unresolvable view format leave it false.
MG_Pipe::MGPipeHandle ResolveWireTextureStorage(MG_Pipe::MGPipeHandle handle, Uint32& level,
Uint32& layer, VkFormat* viewFormat = nullptr, Uint32* layerCount = nullptr);
Uint32& layer, VkFormat* viewFormat = nullptr, Uint32* layerCount = nullptr,
Bool* outsideWindow = nullptr);
// Grows the live image of a handle-keyed resource to `requiredMipLevels`, carrying the
// existing levels' content across with an in-command-buffer copy. The caller
// (GenerateMipmap's record arm) has flushed every pending submission first, so the old
@@ -618,6 +629,26 @@ public:
Bool WasTouchedThisRecording(const TextureResource& resource) const {
return resource.lastRecordingGeneration == m_recordingGeneration;
}
// A write is a use too. See TextureResource::lastRecordingWriteGeneration.
void StampResourceRecordingWrite(TextureResource& resource) const {
resource.lastRecordingGeneration = m_recordingGeneration;
resource.lastRecordingWriteGeneration = m_recordingGeneration;
}
void StampTextureRecordingWrite(MG_State::GLState::ITextureObject* texture);
// Upload-ordering queries (VulkanRenderer::OrderPendingUploadAfterRecording). None of them
// syncs, so nothing pending is uploaded by asking. All answer for the STORAGE texture.
//
// Uploads are recorded into a batch submitted AHEAD of the frame's recording
// (FlushPendingUploads), so an upload into an image the open recording already WROTE lands
// before that write, which then overwrites the newer texels. (An image the recording only
// read keeps the old trade: its earlier reads see the newer texels.)
Bool HasLiveImage(MG_State::GLState::ITextureObject& textureOrView) const;
// The next sync would record client texels into the image that is live now: same shape (a
// shape change mints a new image instead) and content moved since the last sync.
Bool HasPendingTexelUpload(MG_State::GLState::ITextureObject& textureOrView) const;
Bool WasTextureWrittenThisRecording(MG_State::GLState::ITextureObject& textureOrView) const;
// The live resource, or null.
TextureResource* FindTextureResource(MG_State::GLState::ITextureObject& textureOrView);
// Records that this texture is bound to a GL image unit, so its image must carry
// VK_IMAGE_USAGE_STORAGE_BIT. Must be called before NeedsStorageImagePreparation, and
// therefore before the render pass is committed: an image that has to be upgraded is
File diff suppressed because it is too large Load Diff
@@ -147,8 +147,38 @@ namespace MobileGL::MG_Backend::DirectVulkan {
#if MOBILEGL_BUILD_DISAGGREGATED
VkBufferManager& GetWireBufferManager() { return m_bufferManager; }
Bool FlushWirePendingCommandsForTextureUpdate() {
return !HasPendingRecordedWork() || FlushPendingCommands();
// A new texture upload must not race graphics work that samples
// the old texels. No current recording does not imply no in-flight
// submission: finish the recorded work, then wait its whole prefix.
if (HasPendingRecordedWork() && !FlushPendingCommands()) return false;
return WaitForSubmitsUpTo(m_submitCounter, UINT64_MAX);
}
// P7 wave 2 package B3: the submission that will carry whatever is recorded NEXT.
// RetireWireObjects (WireDraw.inc) already tags future objects with exactly this, and
// for the same reason: a draw being set up now is not in any submission yet, so the
// index that names it is one past the counter. Deliberately NOT
// GetSyncPointSubmitIndex(), which answers m_submitCounter when nothing is recorded -
// true for a fence taken at that instant, wrong for work about to be recorded.
Uint64 GetWireNextSubmitIndex() const { return m_submitCounter + 1; }
// P8-D: the wire arm's indirect draw families, issued as vkCmdDraw[Indexed]Indirect[Count]
// from the wire stores MGPipeApplier().VerbIndirectBuffer / VerbIndirectParameterBuffer
// name. `offset` / `countOffset` are byte offsets into those stores; `stride` 0 means
// tightly packed. FALSE = this device cannot issue the COUNT form natively (the monolith
// arm's own condition, see the definition in WireDraw.inc) and the caller must read the
// words on the CPU; TRUE = issued, or declined by name.
Bool DrawWireIndirectNative(GLenum mode, GLenum type, Uint64 offset, GLsizei drawcount, GLsizei stride,
Bool indexed, Bool counted, Uint64 countOffset);
// P11 B2 (T0): what importing a client's AHardwareBuffer as a wire buffer store needs from
// this device. False when the device did not take VK_ANDROID_external_memory_android_
// hardware_buffer at creation (not Android, not advertised, or a monolith device).
struct WireAhbImport {
VkDevice device = VK_NULL_HANDLE;
VkQueue queue = VK_NULL_HANDLE;
Uint32 queueFamily = 0;
VkPhysicalDeviceMemoryProperties memory{};
void* getAhbProperties = nullptr; // PFN_vkGetAndroidHardwareBufferPropertiesANDROID
};
Bool GetWireAhbImport(WireAhbImport& out) const;
#endif
// FrameContext::IRecordingObserver: prepares the frame's timer-query
@@ -316,6 +346,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// signal as soon as the GPU actually finishes, which MC 1.21.5's
// fence-paced ring buffers rely on to recycle their space.
Uint64 GetSyncPointSubmitIndex() const;
// For tests: the per-frame descriptor pools (see DescriptorPoolCensus.h).
DescriptorPoolCensus GetDescriptorPoolCensus() const {
return m_uniformManager ? m_uniformManager->GetDescriptorPoolCensus() : DescriptorPoolCensus{};
}
// Non-blocking: polls outstanding submission fences and reports
// whether every submission up to `submitIndex` has completed.
Bool IsSubmitIndexComplete(Uint64 submitIndex);
@@ -379,6 +413,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool ResolveOcclusionQueryResult(const Vector<Uint32>& slots, Uint64& outSamples);
void RequestSwapchainResize(Uint32 width, Uint32 height);
// eglSwapInterval. Records the request; Present rebuilds the swapchain for it once the
// current frame is on screen, and only when it maps to a different present mode. A
// renderer created without an app window ignores it: its surface (headless, or an
// AImageReader nobody consumes) is never displayed, and FIFO there would block once
// the reader's queue fills.
void SetSwapInterval(Int interval);
// Re-query the surface and report whether the live swapchain no longer matches it
// (size or orientation). This - not a VK_SUBOPTIMAL_KHR result - is what decides a
// rebuild, so a surface the driver merely considers suboptimal cannot thrash.
@@ -405,7 +445,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Int padding[3] = {0, 0, 0};
};
// P7 wave 2-B2, CONTRACT-P7 §5.2 (B'): THE FRONTEND MEMBERS ARE PULL-BUILD ONLY. In a
// disaggregated build the monolith arm blits and generates depth mips with the baked
// modules (WireColorBlit.inc, WireDepthMipmap.inc), so no DirectVulkan object holds a
// ProgramObject, a ShaderObject or a SamplerObject any more - which is what takes
// ProgramObject::{AllocateLifetimeId, AttachShader, Link, ~ProgramObject},
// ShaderObject::{SetShaderSource, Compile, JoinPendingCompile, ReleaseCompileNode,
// DropCompileNode} and SamplerObject::{SetWrapS, SetWrapT, SetWrapR, SetLodRange} out
// of the link ratchet's `p7-magma` bucket (§4.2). The destructor of a SharedPtr member
// is a reference too, so the MEMBER has to go and not only its construction.
//
// The structs stay, empty, so ShutdownBlitResources and the two `= {}` assignments read
// the same on both builds.
struct BlitResources {
#if !MOBILEGL_BUILD_DISAGGREGATED
SharedPtr<MG_State::GLState::ProgramObject> program;
SharedPtr<MG_State::GLState::SamplerObject> nearestSampler;
SharedPtr<MG_State::GLState::SamplerObject> linearSampler;
@@ -413,15 +466,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Int dstRectLocation = -1;
Int surfaceTransformLocation = -1;
Uint32 samplerBinding = 0;
#endif
};
struct DepthMipmapResources {
#if !MOBILEGL_BUILD_DISAGGREGATED
SharedPtr<MG_State::GLState::ProgramObject> program;
Int srcRectLocation = -1;
Int dstRectLocation = -1;
Int surfaceTransformLocation = -1;
Int srcTexelSizeLocation = -1;
Uint32 samplerBinding = 0;
#endif
};
// A single-sample staging image for multisample-resolve blits that also have to change
@@ -467,6 +523,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkImageLayout* trackedLayout = nullptr;
VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;
VkImageAspectFlags aspect = 0;
// Zero for the default framebuffer, which no arm that reads this ever reaches.
VkImageUsageFlags usageFlags = 0;
Uint32 level = 0, layer = 0, layers = 1, levels = 1;
Bool isDefault = false;
// Set when isDefault: the swapchain image index this role resolved to.
@@ -488,18 +546,89 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void ResolveWireDepthStencil(WireImage source, WireImage destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1);
// One aspect of a depth/stencil image through a buffer (source in TRANSFER_SRC_OPTIMAL,
// destination in TRANSFER_DST_OPTIMAL, one format). The returned buffer is named by the
// recorded commands and must be retired with them.
UniquePtr<VkBufferObject> CopyWireAspectThroughBuffer(VkImage srcImage, Uint32 srcLevel, Uint32 srcLayer,
GLint sx, GLint sy, const WireImage& destination,
GLint dx, GLint dy, Uint32 width, Uint32 height,
VkImageAspectFlags aspect, VkFormat format,
Bool mirrorY, const char* fatalDetail);
// A single-sample depth/stencil blit whose rectangles make it a plain copy, performed
// without vkCmdBlitImage (P7 gate 5). False when the shape is not a plain copy.
Bool CopyWireDepthStencilBlit(WireImage& source, WireImage& destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1, VkImageAspectFlags aspect);
// True when the rectangles give a multisample depth/stencil resolve a shape it declines
// (a scale, an X mirror, a Y mirror across two formats); the decline has then been
// logged and its INVALID_OPERATION recorded. BlitWireFramebuffers asks it BEFORE any
// aspect runs, so a call the SHAPE declines leaves every attachment alone - the user
// and the default draw framebuffer alike. The CAPABILITY decline inside
// ResolveWireDepthStencil (a stencil-only resolve without VK_EXT_shader_stencil_export
// on a device without VK_KHR_depth_stencil_resolve) is still decided per aspect, after
// colour ran; rule I (a) accepts that (notes/p7/magma-b2.md §2.10).
Bool DeclineWireDepthStencilResolveShape(const WireImage& source, const WireImage& destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1);
PFN_vkCreateRenderPass2 m_wireCreateRenderPass2 = nullptr;
VkResolveModeFlags m_wireDepthResolveModes = 0;
VkResolveModeFlags m_wireStencilResolveModes = 0;
// P7 wave 2-B2, CONTRACT-P7 §3.2 (`multisample-blit-aspect`): the SECOND arm of the
// multisample depth/stencil resolve, for the device that does not carry
// VK_KHR_depth_stencil_resolve. One baked pass per aspect, sample zero, no frontend
// object - see WireMultisampleResolve.inc. `resolved` must be a single-sample image of
// the source's format and extent; the rect is expressed by the scissor.
Bool ResolveWireDepthStencilWithShader(WireImage source, WireImage resolved,
VkImageAspectFlags aspect, GLint sx0, GLint sy0,
Uint32 width, Uint32 height);
struct WireMultisampleResolveResources;
WireMultisampleResolveResources* m_wireMultisampleResolveResources = nullptr;
void DestroyWireMultisampleResolveResources();
Bool m_wireShaderStencilExport = false;
// VK_EXT_robustness2 (or its KHR promotion) with ONLY nullDescriptor enabled, wire arms
// only: an invalid image unit binds a null storage descriptor (codex closeout finding 2).
Bool m_wireNullDescriptor = false;
// The resolve probe's verdict for this device (WireDepthResolveArm.h,
// WireDepthResolveProbe.h): true when the no-draw render pass was measured leaving its
// target unwritten while the shader control resolved - the shader arm then resolves
// first and the render-pass arm is the fallback (P7 gate 5, g5-msprobe).
Bool m_wirePreferShaderDepthResolve = false;
// P11 B2 (T0): the device took the AHardwareBuffer import extensions at creation, and the
// one entry point the import needs (PFN_vkGetAndroidHardwareBufferPropertiesANDROID).
Bool m_wireAhbImport = false;
void* m_wireGetAhbProperties = nullptr;
// Runs the probe (memoized per device identity) and sets the member above. Called at the end of
// device creation, after ArmPrimGenReroute: it records on m_graphicsQueue.
void ArmWireDepthResolveOrder();
Bool BlitWireColorToDefault(WireImage source, WireImage destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1, GLenum filter);
Bool BlitWireColorImage(WireImage source, WireImage destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1, GLenum filter, Bool mipmap);
// P7 wave 2-B, CONTRACT-P7 §3.2 (`default-color-blit-shape`): the rotated arm. Moves the
// source region into an owned single-sample 2D image of a sampled-float format in the
// same size-compatibility class, then shader-blits THAT to the rotated default
// framebuffer - the shapes BlitWireColorImage itself cannot sample (an integer format, a
// 3D or array source) become shapes it can.
Bool BlitWireColorToDefaultThroughScratch(WireImage source, WireImage destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1,
GLenum filter);
struct WireColorBlitResources;
WireColorBlitResources* m_wireColorBlitResources = nullptr;
void DestroyWireColorBlitResources();
// P7 wave 2-B, CONTRACT-P7 §5.1 (A): the baked depth-mip program (WireDepthMipmap.inc).
// One level of a depth chain, source and destination being two depth subresources of
// the same image; the caller loops. It owns no frontend object, which is what the
// monolith arm's GenerateDepthMipmapWithShader cannot say.
Bool GenerateWireDepthMipLevel(WireImage source, WireImage destination);
struct WireDepthMipmapResources;
WireDepthMipmapResources* m_wireDepthMipmapResources = nullptr;
void DestroyWireDepthMipmapResources();
// P7 wave 2-B (exit gate 3): the readback's own write-visibility barrier, recorded before
// the copy instead of inferred from the image's tracked layout. See WireFramebuffer.inc.
void RecordWireReadbackWriteBarrier();
void ReadWirePixels(GLint x, GLint y, GLsizei width, GLsizei height,
GLenum format, GLenum type, void* pixels);
void GenerateWireMipmap();
@@ -559,7 +688,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
void CollectWireObjects(Uint64 completedSubmit, Bool all = false);
Vector<WireRetiredObjects> m_wireRetiredObjects;
Uint32 m_wirePreparationDepth = 0;
// P8-D: set only while DrawWireIndirectNative is inside SetupDraw, so SetupWireDraw can
// tell the wire arm's own indirect draw from a monolith indirect entry point reached under
// a transport (still the `buffer-legacy-arm` role violation).
Bool m_wireNativeIndirectDraw = false;
void DispatchWireCompute(GLuint x, GLuint y, GLuint z);
// P8-SV: glDispatchComputeIndirect on the wire arm - vkCmdDispatchIndirect from the
// dispatch-indirect store, as the monolith arm issues it (WireDraw.inc).
void DispatchWireComputeIndirect(GLintptr offset);
void RewindWireDescriptorSetsIfDue();
UniquePtr<RenderPassEntry> m_wireDrawPass;
Vector<VkImageView> m_wireDrawViews;
VkPipeline GetOrCreatePipelineWithInput(GLenum mode, const MagmaProgramSource& program,
@@ -616,6 +753,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// was waited or the device was idled); drops their records and
// recycles pooled fences.
void OnSubmitsCompletedUpTo(Uint64 submitIndex);
#if MOBILEGL_BUILD_DISAGGREGATED
// A later fence alone cannot retire earlier submissions. Wait for the
// whole still-live prefix before advancing the completed submit floor.
Bool WaitForSubmitsUpTo(Uint64 submitIndex, Uint64 timeoutNs);
#endif
VkFence AcquirePooledSubmitFence();
void DestroySubmitFencePool();
Bool HasPendingRecordedWork() const;
@@ -628,6 +770,15 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// only), so the presenting path's frames-in-flight pipelining is
// untouched. Returns true when the drain ran.
Bool TryDrainFrameTransients();
// Bounded recording: called at the top of every GL draw and dispatch, before anything
// of that call is recorded. Once the open frame command buffer holds
// MagmaMaxDrawsPerCommandBuffer draws/dispatches, submits it (FlushPendingCommands) so
// the call continues on a fresh one, then waits until at most frames-in-flight + 1
// submissions are outstanding. A driver backs a command buffer with GPU memory it only
// returns when the buffer is freed (Adreno: one 16 KiB mapping per chunk), so a frame
// that is never split grows it without bound - rd12's loading frame records ~1M draws
// into one buffer and runs the process out of vm.max_map_count.
void SplitOversizedRecording();
Vector<SubmitRecord> m_inFlightSubmits;
Vector<VkFence> m_freeSubmitFences;
@@ -639,6 +790,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// aging clocks, while present-less loops still cross a boundary every few
// iterations. Reset in Present.
Uint32 m_drainsSinceLastPresent = 0;
// GL draws/dispatches recorded into the open frame command buffer; reset whenever a
// recording begins (OnFrameCommandRecordingBegan). See SplitOversizedRecording.
Uint32 m_drawsInRecording = 0;
Uint64 m_oversizedRecordingSplits = 0;
NativeWindowType m_window = 0;
void* m_platformDisplay = nullptr;
@@ -658,6 +813,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
// Shutdown() deletes it.
void* m_fallbackImageReader = nullptr;
VulkanRendererConfig m_config;
// Constructed for the app's window rather than a pbuffer (m_window is later filled in
// for the pbuffer fallbacks too, so it cannot tell them apart).
Bool m_presentsToAppWindow = false;
// The m_config.SwapInterval the live swapchain was built for; Present compares the two.
Optional<Int> m_swapchainSwapInterval;
Bool m_swapchainResizeRequested = false;
// Presentation is suspended while the window is zero-area (minimized): the
// swapchain is unusable/out of date, so Present drops frames instead of
@@ -776,6 +936,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
VkDeviceSize countBufferOffset, Uint32 maxDrawCount,
Uint32 stride);
static inline PFNDrawIndexedIndirectCountFunc s_vkCmdDrawIndexedIndirectCount = nullptr;
#if MOBILEGL_BUILD_DISAGGREGATED
// P8-D: vkCmdDrawIndirectCount (same signature, same extension), for the wire arm's
// glMultiDrawArraysIndirectCount. The monolith arm never loads it: it reads that count on
// the CPU (DirectVulkan.cpp's MultiDrawArraysIndirectCount).
static inline PFNDrawIndexedIndirectCountFunc s_vkCmdWireDrawIndirectCount = nullptr;
#endif
// VK_EXT_multi_draw entry points, loaded at device creation when the extension
// (and its multiDraw feature) is enabled; null otherwise.
static inline PFN_vkCmdDrawMultiEXT s_vkCmdDrawMultiEXT = nullptr;
@@ -1759,8 +1925,28 @@ namespace MobileGL::MG_Backend::DirectVulkan {
MG_State::GLState::ITextureObject& texture, Uint32 mipLevel,
Uint32 depthSlice, const VkClearValue& clearValue,
VkImageLayout finalLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
// Records the texture's queued clears. By default the texture is synced first, which
// uploads any pending texels BEFORE the clear - right only for texels written before it.
// `uploadPendingTexels = false` leaves them pending (OrderPendingUploadAfterRecording).
Bool MaterializePendingClearForTexture(VkCommandBuffer commandBuffer,
MG_State::GLState::ITextureObject& texture);
MG_State::GLState::ITextureObject& texture,
Bool uploadPendingTexels = true);
// Texture uploads run in a batch submitted ahead of the frame's recording, so the upload
// the next sync of `texture` records would land BEFORE a write already recorded into the
// same image (a render, clear, blit or copy), and before a clear still queued for it (such
// a clear predates the texels: SettleTexelsBeforeQueuedClear). When either holds, this
// records the queued clear without the texels and submits the recording, so the upload
// follows both. An image the recording only READ keeps the old trade: no submit, and its
// earlier reads see the newer texels. It may end the render pass and replace
// frame.commandBuffer (recording again on return), so call it only where the caller holds
// no render-pass or command-buffer state across it. Returns whether it submitted.
Bool OrderPendingUploadAfterRecording(FrameContext::FrameData& frame,
MG_State::GLState::ITextureObject& texture);
// Called for each texture a glClear/glClearBuffer* is about to queue a clear on: uploads
// whatever texels are still pending (and mints the image if there is none), so that any
// texel pending when the clear is materialized was written after it.
void SettleTexelsBeforeQueuedClear(const MG_State::GLState::FramebufferObject& framebuffer,
const MG_State::GLState::FramebufferAttachmentObject& attachment);
// The multisample arm of the above. Split out rather than branched inline because it
// shares none of the transfer path: a multisample image carries no TRANSFER_DST usage, so
// neither the TRANSFER_DST transition nor vkCmdClearColorImage is legal on one.
@@ -1783,7 +1969,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Bool MaterializePendingDepthStencilClearForDefaultFramebuffer(
VkCommandBuffer commandBuffer, const MG_State::GLState::FramebufferAttachmentObject& attachment,
const ClearAttachmentPayload& payload);
#if !MOBILEGL_BUILD_DISAGGREGATED
// P7 wave 2-B2 (CONTRACT-P7 §5.2, (B')): the hidden blit program's pipeline. A
// disaggregated build has no hidden program on either arm, so this has neither a body
// nor a caller there.
VkPipeline GetOrCreateBlitPipeline(const RenderPassEntry& renderPassEntry);
#endif
Bool GenerateDepthMipmapWithShader(FrameContext::FrameData& frame,
MG_State::GLState::ITextureObject& texture,
VkTextureManager::TextureResource& resource,
@@ -43,6 +43,112 @@
return BlitWireColorImage(source, destination, sx0, sy0, sx1, sy1, dx0, dy0, dx1, dy1, filter, false);
}
// P7 wave 2-B (CONTRACT-P7 §3.2, `default-color-blit-shape`). The shader blit below can only
// SAMPLE its source: sampler2D, a float result, and one of two 2D view types. A framebuffer
// whose colour attachment is RGBA8UI, or a layer of a 3D texture, is none of those, and on a
// rotated Android surface there is no other arm - an axis-aligned vkCmdBlitImage cannot
// express a quarter turn. So the source region becomes an owned 2D image first, through
// vkCmdCopyImage, which cares about the texel BLOCK SIZE and not about how the block is
// interpreted; the scratch then takes the shader blit as an ordinary 2D float source.
//
// REINTERPRETATION IS THE POINT FOR THE INTEGER CASE, and GL agrees: glBlitFramebuffer
// between an integer and a normalized format is INVALID_OPERATION (GL 4.6 core 18.3.1), so
// there is no defined result to be wrong about - only a shape that used to end the session
// and now produces the source's bits. For a float source of the wrong VIEW TYPE the copy is
// a plain slice extraction and the result is exactly right.
Bool VulkanRenderer::BlitWireColorToDefaultThroughScratch(WireImage source, WireImage destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1, GLenum filter) {
if (!destination.isDefault || source.isDefault) return false;
if (source.samples != VK_SAMPLE_COUNT_1_BIT || destination.samples != VK_SAMPLE_COUNT_1_BIT)
return false;
if (!source.image || !destination.image || source.image == destination.image) return false;
const Int64 width = std::abs(static_cast<Int64>(sx1) - sx0);
const Int64 height = std::abs(static_cast<Int64>(sy1) - sy0);
if (!width || !height || dx0 == dx1 || dy0 == dy1) return true;
if (std::min(sx0, sx1) < 0 || std::min(sy0, sy1) < 0 ||
static_cast<Uint64>(std::max(sx0, sx1)) > source.extent.width ||
static_cast<Uint64>(std::max(sy0, sy1)) > source.extent.height) return false;
// A compressed source cannot be a colour attachment, so it cannot be a framebuffer blit's
// source either - but the block arithmetic below would silently produce nonsense for one,
// and "cannot happen" is the reason the guards this package retires were written as
// asserts. The size-compatibility classes are Vulkan's own (VkFormat texel block size);
// the destination format is whatever the swapchain uses and is always sampled-float.
if (vkuFormatIsCompressed(source.format)) return false;
VkFormat scratchFormat = VK_FORMAT_UNDEFINED;
switch (vkuFormatTexelBlockSize(source.format)) {
case 1: scratchFormat = VK_FORMAT_R8_UNORM; break;
case 2: scratchFormat = VK_FORMAT_R8G8_UNORM; break;
case 4: scratchFormat = VK_FORMAT_R8G8B8A8_UNORM; break;
case 8: scratchFormat = VK_FORMAT_R16G16B16A16_SFLOAT; break;
case 16: scratchFormat = VK_FORMAT_R32G32B32A32_SFLOAT; break;
default: return false;
}
VkFormatProperties scratchProperties{};
vkGetPhysicalDeviceFormatProperties(m_physicalDevice.handle, scratchFormat, &scratchProperties);
const VkFormatFeatureFlags scratchRequired = VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT |
(filter == GL_LINEAR ? VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT : 0);
if ((scratchProperties.optimalTilingFeatures & scratchRequired) != scratchRequired) return false;
auto scratch = MakeUnique<VkTextureManager::TextureResource>();
VkImageCreateInfo info{VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO};
info.imageType = VK_IMAGE_TYPE_2D;
info.format = scratchFormat;
info.extent = {static_cast<Uint32>(width), static_cast<Uint32>(height), 1};
info.mipLevels = info.arrayLayers = 1;
info.samples = VK_SAMPLE_COUNT_1_BIT;
info.tiling = VK_IMAGE_TILING_OPTIMAL;
info.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT;
info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
VmaAllocationCreateInfo allocation{};
allocation.usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
allocation.requiredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
if (vmaCreateImage(m_allocator, &info, &allocation, &scratch->image, &scratch->allocation,
nullptr) != VK_SUCCESS)
return false;
scratch->format = scratchFormat;
scratch->extent = {info.extent.width, info.extent.height};
scratch->mipLevels = scratch->arrayLayers = scratch->depth = 1;
scratch->aspect = VK_IMAGE_ASPECT_COLOR_BIT;
WireImage staged{};
staged.image = scratch->image;
staged.format = scratchFormat;
staged.aspect = VK_IMAGE_ASPECT_COLOR_BIT;
staged.extent = scratch->extent;
staged.trackedLayout = &scratch->layout;
m_textureManager->FlushPendingUploads();
TransitionWireImage(source, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
TransitionWireImage(staged, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
VkImageCopy copy{};
// A 3D source carries its slice on z and always names layer 0; every other source
// carries it on the layer axis, which is the same split ResolveWireImage records.
copy.srcSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, source.level, source.is3D ? 0u : source.layer, 1};
copy.srcOffset = {std::min(sx0, sx1), std::min(sy0, sy1),
source.is3D ? static_cast<Int32>(source.layer) : 0};
copy.dstSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1};
copy.dstOffset = {0, 0, 0};
copy.extent = {static_cast<Uint32>(width), static_cast<Uint32>(height), 1};
vkCmdCopyImage(m_frameContext.GetCurrent().commandBuffer, source.image,
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, staged.image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &copy);
staged.layout = scratch->layout;
// The scratch holds the region at its own origin, so a reversed source rectangle
// becomes a reversed rectangle OVER THE SCRATCH - the multisample resolve arm's shape.
const GLint rx0 = sx1 > sx0 ? 0 : static_cast<GLint>(width);
const GLint rx1 = sx1 > sx0 ? static_cast<GLint>(width) : 0;
const GLint ry0 = sy1 > sy0 ? 0 : static_cast<GLint>(height);
const GLint ry1 = sy1 > sy0 ? static_cast<GLint>(height) : 0;
const Bool blitted = BlitWireColorImage(staged, destination, rx0, ry0, rx1, ry1,
dx0, dy0, dx1, dy1, filter, false);
// Retired whether or not the blit took: the copy above is already recorded against this
// image, so the submission that consumes it owns the storage either way.
WireRetiredObjects retired{};
retired.textures.push_back(Move(scratch));
RetireWireObjects(Move(retired));
return blitted;
}
Bool VulkanRenderer::BlitWireColorImage(WireImage source, WireImage destination,
GLint sx0, GLint sy0, GLint sx1, GLint sy1,
GLint dx0, GLint dy0, GLint dx1, GLint dy1, GLenum filter, Bool mipmap) {
@@ -234,7 +340,14 @@
alloc.descriptorSetCount = 1;
alloc.pSetLayouts = &cache.descriptorLayout;
VK_VERIFY(vkAllocateDescriptorSets(m_device, &alloc, &descriptorSet));
const VkDescriptorImageInfo sampled{filter == GL_LINEAR ? cache.linear : cache.nearest,
// GL_LINEAR filters only a STRETCHED blit; an unscaled one (mirrored or not, on any of the
// four transforms) is a copy and samples NEAREST, which is exact where linear filtering
// at a quantised coordinate is not (WireColorBlitFilter.h). The format check above still
// asks for SAMPLED_IMAGE_FILTER_LINEAR whenever GL said GL_LINEAR, so which blits reach
// this pass is unchanged.
const VkFilter samplerFilter = WireColorBlitFilterFor({sx0, sy0, sx1, sy1}, {dx0, dy0, dx1, dy1},
filter, transform);
const VkDescriptorImageInfo sampled{samplerFilter == VK_FILTER_LINEAR ? cache.linear : cache.nearest,
sourceView, sourceLayout};
VkWriteDescriptorSet write{VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET};
write.dstSet = descriptorSet;
@@ -0,0 +1,66 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WireColorBlitFilter.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Includes.h>
#include "../VkIncludes.h"
// P7 g3-blit (g3det, gate 3): WHICH SAMPLER THE WIRE ARM'S SHADER COLOUR BLIT READS ITS SOURCE
// THROUGH.
//
// GL applies a blit's `filter` only where the image is STRETCHED (GL 4.6 core 18.3.1). A blit whose
// source and destination rectangles are the same size is a copy, whatever `filter` says, and so is
// a mirrored one. The wire arm's shader blit (WireColorBlit.inc, BlitWireColorImage) draws a quad
// and samples `textureLod(sourceImage, texCoord, 0)` at interpolated coordinates, and it used the
// LINEAR sampler for every GL_LINEAR blit. At 1:1 every fragment centre maps to a texel centre, so
// in exact arithmetic linear filtering reads the same single texel. Adreno 830 quantises the
// coordinate to sub-texel precision: a fragment that lands a fraction off the centre blends in a
// small weight of a neighbour. minecraft-1.21.1-neoforge-create-instancing-in-world ends with a
// 1:1 GL_LINEAR glBlitFramebuffer into FB 0, and on the wire arm 712 high-contrast pixels moved by
// 1 LSB toward a neighbour. The monolith's vkCmdBlitImage copied them exactly
// (~/w7/logs/g3det/VERDICT.md section 3). NEAREST within half a texel of the centre is exact, so an
// unscaled blit takes NEAREST. Lavapipe samples texel centres exactly, so no host picture can see
// the difference; this pure half is what the unit test pins
// (MG_Test/Pipeline/PipelineQuirkTest.cpp, WireColorBlitSamples*).
//
// THE SURFACE TRANSFORM DOES NOT TURN A COPY INTO A STRETCH. BlitWireColorImage applies the
// quarter and half turns in NDC, over a viewport of the native image, and for a quarter turn it
// takes the logical extent as the native extent swapped. A native pixel then has exactly the pitch
// of a logical one, and the rotation maps pixel centres onto pixel centres. So the scale test is on
// GL's own (logical) rectangles for identity and all three rotations. BlitWireColorImage declines
// any other transform before it chooses a sampler; for one, this answers with the filter's own
// mapping, as before this package.
#if MOBILEGL_BUILD_DISAGGREGATED
namespace MobileGL::MG_Backend::DirectVulkan {
// A glBlitFramebuffer rectangle as GL names it: two corners, either order.
struct WireBlitRect {
GLint x0 = 0, y0 = 0, x1 = 0, y1 = 0;
};
// Same extent on both axes, orientation aside. The extents are taken in 64 bits: GL accepts any
// GLint corner, and INT_MAX - INT_MIN does not fit in one.
inline Bool WireBlitIsUnscaled(const WireBlitRect& source, const WireBlitRect& destination) {
const auto extent = [](GLint a, GLint b) {
return std::abs(static_cast<Int64>(b) - static_cast<Int64>(a));
};
return extent(source.x0, source.x1) == extent(destination.x0, destination.x1) &&
extent(source.y0, source.y1) == extent(destination.y0, destination.y1);
}
inline VkFilter WireColorBlitFilterFor(const WireBlitRect& source, const WireBlitRect& destination,
GLenum filter, VkSurfaceTransformFlagBitsKHR transform) {
if (filter != GL_LINEAR) return VK_FILTER_NEAREST;
const Bool expressed = transform == VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR ||
transform == VK_SURFACE_TRANSFORM_ROTATE_90_BIT_KHR ||
transform == VK_SURFACE_TRANSFORM_ROTATE_180_BIT_KHR ||
transform == VK_SURFACE_TRANSFORM_ROTATE_270_BIT_KHR;
if (expressed && WireBlitIsUnscaled(source, destination)) return VK_FILTER_NEAREST;
return VK_FILTER_LINEAR;
}
} // namespace MobileGL::MG_Backend::DirectVulkan
#endif
@@ -1,6 +1,11 @@
// Generated from WireColorBlit.vert / WireColorBlit.frag with:
// glslangValidator -V --target-env vulkan1.0 -o <stage>.spv <source>
// Validated with spirv-val --target-env vulkan1.0. Do not edit the words by hand.
// Generated from WireColorBlit.vert / WireColorBlit.frag (Vulkan 1.0 / SPIR-V 1.0) by
// scripts/bake_internal_shaders.py, which runs MG_Test's BakedInternalShadersTest with the
// IN-TREE glslang and splices the words it emits into the arrays below. Do not edit the words
// by hand: MOBILEGL_BAKED_INTERNAL_SHADERS (the same test, run as a gate) recompiles every
// source in its table and reds on the first word that disagrees.
//
// These words were hand-run through glslangValidator before that gate existed, and the gate
// reproduces them exactly - which is the one thing nobody could have known beforehand.
#pragma once
static constexpr Uint32 kWireColorBlitVertexSpirv[] = {
@@ -0,0 +1,172 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WireDeclineTally.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Config.h>
#if MOBILEGL_BUILD_DISAGGREGATED
#include <Includes.h>
#include <MG_Util/Debug/Log.h>
#include <atomic>
namespace MobileGL::MG_Backend::DirectVulkan {
// P7 wave 2 package B3: the per-site tally behind CONTRACT-P7 §0 rule I.
//
// Shape is PipeApplier.h's ServerVerbSink tallies, and for the same reason (R-16: a probe may
// not arm against a stub). "The retrace passed" and "the retrace silently dropped nine draws"
// produce the same exit code and the same log, so the lane has to assert the number that only
// this path can move.
//
// Counters are process-wide and monotone. On the spawn arm they live in the SERVER process,
// which is where the draw path runs - a client-side reader would always see zero, so the
// two-process case asserts pixels and reads these from the server's own dump.
enum class WireDeclineSite : Uint32 {
#define MGL_WIRE_DECLINE(name) name,
#include "WireDeclines.def"
#undef MGL_WIRE_DECLINE
Count
};
inline const char* WireDeclineSiteName(WireDeclineSite site) {
switch (site) {
#define MGL_WIRE_DECLINE(name) \
case WireDeclineSite::name: \
return #name;
#include "WireDeclines.def"
#undef MGL_WIRE_DECLINE
case WireDeclineSite::Count:
break;
}
return "Unnamed";
}
class WireDeclineTally {
public:
static constexpr SizeT kSiteCount = static_cast<SizeT>(WireDeclineSite::Count);
static void Count(WireDeclineSite site) {
Slots()[static_cast<SizeT>(site)].fetch_add(1, std::memory_order_relaxed);
Total().fetch_add(1, std::memory_order_relaxed);
}
static Uint64 Get(WireDeclineSite site) {
return Slots()[static_cast<SizeT>(site)].load(std::memory_order_relaxed);
}
static Uint64 TotalDeclines() { return Total().load(std::memory_order_relaxed); }
// The memory half (package M1's thread): a refused upload does not consume its
// PendingUpload entry, so the entry and the level shadow behind it stay alive. These are
// GAUGES, not counters - the last observed live depth, sampled where the refusal happens -
// so a monotone climb across frames is the signature that a refusal is also a leak.
static void SetPendingGauge(Uint64 entries, Uint64 bytes) {
PendingEntries().store(entries, std::memory_order_relaxed);
PendingBytes().store(bytes, std::memory_order_relaxed);
Uint64 peak = PeakPendingBytes().load(std::memory_order_relaxed);
while (bytes > peak && !PeakPendingBytes().compare_exchange_weak(peak, bytes, std::memory_order_relaxed)) {
}
}
// The frame-serial floor's own counter. An "unsound serial complete" is an advance of
// VkBufferManager's completed-frame-serial floor to a serial that ANOTHER submission still
// in flight also carries - i.e. the renderer telling the buffer manager "frame N-1 is done"
// while the submission holding frame N-1's buffer copies is still executing. It is what
// lets a streamed glBufferSubData take the unordered host path and tear one draw's vertex
// range. With a provable floor it is unreachable, so this counter is the red-once's reading:
// non-zero on the wire arms before the fix, zero after, zero on monolith either way.
static void CountUnsoundSerialComplete() {
UnsoundSerial().fetch_add(1, std::memory_order_relaxed);
}
static Uint64 UnsoundSerialCompleteEvents() { return UnsoundSerial().load(std::memory_order_relaxed); }
static Uint64 PendingUploadEntries() { return PendingEntries().load(std::memory_order_relaxed); }
static Uint64 PendingUploadBytes() { return PendingBytes().load(std::memory_order_relaxed); }
static Uint64 PeakPendingUploadBytes() { return PeakPendingBytes().load(std::memory_order_relaxed); }
static void Reset() {
for (SizeT i = 0; i < kSiteCount; ++i) Slots()[i].store(0, std::memory_order_relaxed);
Total().store(0, std::memory_order_relaxed);
UnsoundSerial().store(0, std::memory_order_relaxed);
PendingEntries().store(0, std::memory_order_relaxed);
PendingBytes().store(0, std::memory_order_relaxed);
PeakPendingBytes().store(0, std::memory_order_relaxed);
}
// One line per site that moved, plus the pending gauges. Called at session teardown and
// from the forced-condition harness; silent when nothing declined, so a green lane stays
// quiet and a lane that dropped a draw cannot.
static void Dump(const char* tag) {
// The floor counter prints unconditionally: "zero unsound advances" is the reading the
// red-once needs, and an absent line cannot be told from a lane that never ran.
MGLOG_I("MGWIRE-FLOOR[%s] unsoundSerialComplete=%llu", tag,
static_cast<unsigned long long>(UnsoundSerialCompleteEvents()));
if (TotalDeclines() == 0 && PeakPendingUploadBytes() == 0) return;
MGLOG_I("MGWIRE-DECLINES[%s] total=%llu pendingEntries=%llu pendingBytes=%llu peakPendingBytes=%llu", tag,
static_cast<unsigned long long>(TotalDeclines()),
static_cast<unsigned long long>(PendingUploadEntries()),
static_cast<unsigned long long>(PendingUploadBytes()),
static_cast<unsigned long long>(PeakPendingUploadBytes()));
for (SizeT i = 0; i < kSiteCount; ++i) {
const Uint64 value = Slots()[i].load(std::memory_order_relaxed);
if (!value) continue;
MGLOG_I("MGWIRE-DECLINES[%s] %s=%llu", tag, WireDeclineSiteName(static_cast<WireDeclineSite>(i)),
static_cast<unsigned long long>(value));
}
}
private:
static std::atomic<Uint64>* Slots() {
static std::atomic<Uint64> slots[kSiteCount]{};
return slots;
}
static std::atomic<Uint64>& Total() {
static std::atomic<Uint64> total{0};
return total;
}
static std::atomic<Uint64>& PendingEntries() {
static std::atomic<Uint64> value{0};
return value;
}
static std::atomic<Uint64>& PendingBytes() {
static std::atomic<Uint64> value{0};
return value;
}
static std::atomic<Uint64>& PeakPendingBytes() {
static std::atomic<Uint64> value{0};
return value;
}
static std::atomic<Uint64>& UnsoundSerial() {
static std::atomic<Uint64> value{0};
return value;
}
};
// Every decline goes through this: the tally moves ALWAYS, the log line is once per site.
// `MGL_WIRE_DECLINE_AT(Site, "why", ...); return false;` is the statement pair that replaces
// a bare `return false;` (the few sites that call WireDeclineTally::Count directly stand
// under an MGLOG_W/E of their own; scripts/ci/wire_declines_audit.py holds them to that).
//
// A DECLINE IS NOT ALWAYS A SKIPPED DRAW, so the line does not say "draw". The Tex*, Shape* and
// Upload* rows are counted per texture SYNC: SyncTextureResourceByHandle is reached from the
// draw/dispatch texture preparation, but also from the descriptor resolver, framebuffer
// attachments, CopyTexSubImage, mipmap generation and CopyImageSubData - each of which Fatals
// right after a failed sync - and from texture readback, which falls back to the unbacked
// shadow. A draw that samples three unsyncable textures counts three; a failed attachment
// sync counts once and is then a Fatal, not a skipped draw.
#define MGL_WIRE_DECLINE_AT(site, fmt, ...) \
do { \
::MobileGL::MG_Backend::DirectVulkan::WireDeclineTally::Count( \
::MobileGL::MG_Backend::DirectVulkan::WireDeclineSite::site); \
MGLOG_E_ONCE("Magma wire decline [" #site "]: " fmt, ##__VA_ARGS__); \
} while (0)
} // namespace MobileGL::MG_Backend::DirectVulkan
#endif // MOBILEGL_BUILD_DISAGGREGATED
@@ -0,0 +1,128 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WireDeclines.def
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// P7 wave 2 package B3. THE NAMES OF EVERY WAY THE MAGMA WIRE ARM CAN DECLINE.
//
// A DECLINE IS NOT ALWAYS A SKIPPED DRAW. The SetupWireDraw / DispatchWireCompute rows are one
// per refused draw or dispatch; the Tex* / Shape* / Upload* rows are counted per texture SYNC,
// which the draw path reaches once per sampled texture and the descriptor resolver, framebuffer
// attachments, CopyTexSubImage, mipmap generation, CopyImageSubData and texture readback reach
// on their own (WireDeclineTally.h, above MGL_WIRE_DECLINE_AT).
//
// CONTRACT-P7 §0 rule I says a refusal on the wire arm is either a named decline or a named
// Fatal. Before B3 the draw path had a third shape the rule does not admit: a bare
// `return false` with no log, no counter and no census entry - roughly twenty of them in
// SetupWireDraw alone, plus the texture-sync chain behind PrepareWireTextureResources. A draw
// that took one of those exits left NO trace anywhere, which is precisely why the OpenRA
// device divergence survived four device windows with client and server logs that were
// LINE-IDENTICAL between a passing and a failing run.
//
// Each row below is one such exit. The name is what the decline logs and what the tally
// counts, so "which exit did this frame take" is answerable from a counter dump instead of
// from a bisect. Rows are append-only: the integration lane asserts specific counters moved.
//
// EVERY ROW HAS A SITE AND EVERY SITE HAS A LINE. A row with no site is a name that can never
// be counted - it reads as "this cannot happen here" when it only means "nobody wired it up" -
// and a site with no line puts the reader back on a bisect. Sites reached through a macro
// (MGL_WIRE_DECLINE_AT) get both at once; the few that use WireDeclineTally::Count directly do
// so ONLY because the statement above them already carries the same reason as an _ONCE log.
// scripts/ci/wire_declines_audit.py holds both halves.
//
// MGL_WIRE_DECLINE(Name)
// ---- SetupWireDraw, in statement order -------------------------------------------------
MGL_WIRE_DECLINE(ProgramLinkOrSpirv)
MGL_WIRE_DECLINE(PointSizeCapability)
MGL_WIRE_DECLINE(TextureResourcesUnprepared)
MGL_WIRE_DECLINE(VertexFormatUnsupported)
MGL_WIRE_DECLINE(VertexAttributeWindow)
MGL_WIRE_DECLINE(VertexStreamCountZero)
MGL_WIRE_DECLINE(VertexStreamRead)
MGL_WIRE_DECLINE(VertexStreamConverter)
MGL_WIRE_DECLINE(VertexTransientUpload)
MGL_WIRE_DECLINE(IndexViewMissing)
MGL_WIRE_DECLINE(IndexTypeWidth)
MGL_WIRE_DECLINE(IndexWindow)
MGL_WIRE_DECLINE(IndexStreamRead)
MGL_WIRE_DECLINE(IndexTransientUpload)
MGL_WIRE_DECLINE(IndexSliceInvalid)
MGL_WIRE_DECLINE(UniformBufferBinding)
MGL_WIRE_DECLINE(NoAttachments)
MGL_WIRE_DECLINE(PipelineUnavailable)
MGL_WIRE_DECLINE(CurrentAttributePayload)
MGL_WIRE_DECLINE(CurrentAttributeUpload)
MGL_WIRE_DECLINE(BeginRenderPass)
// ---- DispatchWireCompute (and DispatchWireComputeIndirect, P8-SV) -------------------------
MGL_WIRE_DECLINE(ComputeProgramLinkOrSpirv)
MGL_WIRE_DECLINE(ComputeTextureResourcesUnprepared)
MGL_WIRE_DECLINE(ComputePipelineUnavailable)
MGL_WIRE_DECLINE(ComputeUniformBufferBinding)
// ---- UniformManager::PrepareWireTextureResources ----------------------------------------
MGL_WIRE_DECLINE(DescriptorDeclined)
MGL_WIRE_DECLINE(SamplerLocationAlias)
MGL_WIRE_DECLINE(SamplerUnitRange)
// ---- VkTextureManager::SyncTextureResourceByHandle --------------------------------------
MGL_WIRE_DECLINE(TexResolveStorage)
MGL_WIRE_DECLINE(TexViewFormat)
MGL_WIRE_DECLINE(TexNoApplierRecord)
MGL_WIRE_DECLINE(TexRecordDeadOrStale)
MGL_WIRE_DECLINE(TexShapeSync)
MGL_WIRE_DECLINE(TexUploadPendingLevels)
MGL_WIRE_DECLINE(TexStorageUsage)
// ---- VkTextureManager::UploadPendingWireLevels ------------------------------------------
// The five `incomplete`/false exits the investigation named, plus the two that leave an
// entry pending WITHOUT failing (a `continue`) - those are the ones that can starve a draw
// of its texels while still reporting success, so they are counted separately.
MGL_WIRE_DECLINE(UploadLevelBeyondChain)
MGL_WIRE_DECLINE(UploadLevelExtentMismatch)
MGL_WIRE_DECLINE(UploadEmptyLevel)
MGL_WIRE_DECLINE(UploadConversion)
MGL_WIRE_DECLINE(UploadTexelArithmetic)
MGL_WIRE_DECLINE(UploadRegionOutOfBounds)
MGL_WIRE_DECLINE(UploadIncomplete)
MGL_WIRE_DECLINE(UploadFlushFailed)
// ---- VkTextureManager::SyncWireTextureShape ---------------------------------------------
MGL_WIRE_DECLINE(ShapeBufferBacked)
MGL_WIRE_DECLINE(ShapeViewOfStorage)
MGL_WIRE_DECLINE(ShapeNoLevels)
MGL_WIRE_DECLINE(ShapeNoVulkanShape)
MGL_WIRE_DECLINE(ShapeNoVkFormat)
MGL_WIRE_DECLINE(ShapeSampleCount)
MGL_WIRE_DECLINE(ShapePreserveFlushFailed)
MGL_WIRE_DECLINE(ShapeImageFlagsUnsupported)
MGL_WIRE_DECLINE(ShapeCreateImageFailed)
// Appended by the B3 review round: the preserve-copy arm of a reallocation was the one exit in
// SyncWireTextureShape still leaving no trace at all.
MGL_WIRE_DECLINE(ShapePreserveCopyFailed)
// ---- VulkanRenderer::GenerateWireMipmap (P8-F) -----------------------------------------
// `mipmap-shader-format-or-shape` was a Fatal on this arm and a decline on the monolith arm; it
// is the monolith's decline now. Counted once per refused glGenerateMipmap, whose chain is left
// as it was.
MGL_WIRE_DECLINE(MipmapShaderFormatOrShape)
// ---- VulkanRenderer::DrawWireIndirectNative (P8-D), in statement order -----------------------
MGL_WIRE_DECLINE(IndirectBufferUnbound)
MGL_WIRE_DECLINE(IndirectCommandRange)
MGL_WIRE_DECLINE(IndirectCountRange)
MGL_WIRE_DECLINE(IndirectIndexStore)
// ---- P8-SV -------------------------------------------------------------------------------
// VulkanRenderer::GenerateWireMipmap: a combined depth/stencil (or stencil-only) chain, which no
// arm generates (CONTRACT-P7 §3.2). It logged but was never counted, so MGWIRE-DECLINES could
// not see it. Counted once per refused glGenerateMipmap, whose chain is left as it was.
MGL_WIRE_DECLINE(MipmapDepthStencilAspect)
// VulkanRenderer::DispatchWireComputeIndirect, in statement order: the dispatch-indirect store
// is missing, or the three group-count words lie outside it or are misaligned.
MGL_WIRE_DECLINE(DispatchIndirectBufferUnbound)
MGL_WIRE_DECLINE(DispatchIndirectRange)
@@ -0,0 +1,28 @@
#version 450
// The monolith arm's kDepthMipmapFragmentShaderSource (VulkanRenderer.cpp), ported to the
// baked pipeline: the same 2x2 box of texelFetches, the same clamp, the same average, the
// same gl_FragDepth. It is a PORT and not a rewrite on purpose - CONTRACT-P7 §3.2 retires a
// refusal by making the wire arm observably the monolith's, so the two arms have to agree
// texel for texel, including where they read.
//
// uSrcTexelSize arrives on the push constant rather than in a uniform block: the wire arm
// owns no GL program, so there is no default-block uniform to write and no global UBO to
// bind. That is the whole of OQ-13's narrow half for this program.
layout(push_constant) uniform DepthMipmapRect {
vec4 srcRect;
vec4 dstRect;
int surfaceTransform;
ivec2 srcTexelSize;
} rect;
layout(set = 0, binding = 0) uniform sampler2D sourceImage;
layout(location = 0) in vec2 texCoord;
void main() {
ivec2 srcBase = ivec2(texCoord * vec2(rect.srcTexelSize));
ivec2 srcMax = rect.srcTexelSize - ivec2(1);
float depth0 = texelFetch(sourceImage, clamp(srcBase, ivec2(0), srcMax), 0).r;
float depth1 = texelFetch(sourceImage, clamp(srcBase + ivec2(1, 0), ivec2(0), srcMax), 0).r;
float depth2 = texelFetch(sourceImage, clamp(srcBase + ivec2(0, 1), ivec2(0), srcMax), 0).r;
float depth3 = texelFetch(sourceImage, clamp(srcBase + ivec2(1, 1), ivec2(0), srcMax), 0).r;
gl_FragDepth = 0.25 * (depth0 + depth1 + depth2 + depth3);
}
@@ -0,0 +1,328 @@
// Included by VulkanRenderer.cpp inside the backend namespace, after WireColorBlit.inc.
// The native depth mip owns no frontend object and never invokes the GL shader compiler.
//
// WHY THIS FILE EXISTS AT ALL. vkCmdBlitImage is the ordinary way to halve a mip, and for a
// depth format it is often not available: VK_FORMAT_FEATURE_BLIT_DST is optional for depth,
// and the monolith arm has carried a shader fallback for exactly that case since before the
// split (VulkanRenderer.cpp's GenerateDepthMipmapWithShader). The wire arm had no such
// fallback, so every depth mip the device could not blit died as
// `Magma:mipmap-shader-format-or-shape`, and a DEPTH|STENCIL texture died one guard earlier
// as `Magma:depth-stencil-mipmap@P7` whether or not it could be blitted.
//
// The monolith fallback is not reusable here and that is the whole of CONTRACT-P7 §5.1 (A):
// it builds a GL ProgramObject, two GL ShaderObjects and a GL sampler, binds through the
// uniform manager's global UBO, and writes uSrcTexelSize as a default-block uniform - four
// frontend objects and a GL compiler on the server's apply thread. The baked program below is
// the same shader with the same arithmetic, driven by a push constant and its own descriptor
// set, which is what lets the server generate a depth mip while owning nothing of the client.
#if MOBILEGL_BUILD_DISAGGREGATED
#include "WireDepthMipmapSpirv.h"
struct VulkanRenderer::WireDepthMipmapResources {
struct Target {
VkFormat format = VK_FORMAT_UNDEFINED;
VkImageLayout layout = VK_IMAGE_LAYOUT_UNDEFINED;
VkRenderPass renderPass = VK_NULL_HANDLE;
VkPipeline pipeline = VK_NULL_HANDLE;
};
VkShaderModule vertex = VK_NULL_HANDLE, fragment = VK_NULL_HANDLE;
VkDescriptorSetLayout descriptorLayout = VK_NULL_HANDLE;
VkPipelineLayout pipelineLayout = VK_NULL_HANDLE;
VkSampler nearest = VK_NULL_HANDLE;
Vector<Target> targets;
};
void VulkanRenderer::DestroyWireDepthMipmapResources() {
if (!m_wireDepthMipmapResources) return;
// Shutdown calls this after vkDeviceWaitIdle and after all per-submit
// descriptor pools/framebuffers/views have been collected.
auto& cache = *m_wireDepthMipmapResources;
for (const auto& target : cache.targets) {
vkDestroyPipeline(m_device, target.pipeline, nullptr);
vkDestroyRenderPass(m_device, target.renderPass, nullptr);
}
vkDestroySampler(m_device, cache.nearest, nullptr);
vkDestroyPipelineLayout(m_device, cache.pipelineLayout, nullptr);
vkDestroyDescriptorSetLayout(m_device, cache.descriptorLayout, nullptr);
vkDestroyShaderModule(m_device, cache.vertex, nullptr);
vkDestroyShaderModule(m_device, cache.fragment, nullptr);
delete m_wireDepthMipmapResources;
m_wireDepthMipmapResources = nullptr;
}
// ONE LEVEL, source -> destination, both depth subresources of the same image. The shape
// mirrors BlitWireColorImage deliberately: the same lazily built module pair, the same
// per-format render pass/pipeline cache, the same per-submit descriptor pool, and the
// same GENERAL-on-both-sides layout rule for a same-image pass, so the two native passes
// cannot disagree about how an image is handed between them.
Bool VulkanRenderer::GenerateWireDepthMipLevel(WireImage source, WireImage destination) {
const Bool sameImage = source.image == destination.image;
if (!source.image || !destination.image) return false;
if (sameImage && source.level == destination.level && source.layer == destination.layer)
return false;
if (source.isDefault || destination.isDefault) return false;
if (source.samples != VK_SAMPLE_COUNT_1_BIT || destination.samples != VK_SAMPLE_COUNT_1_BIT)
return false;
if (source.is3D || destination.is3D) return false;
// P8-SV: THE PASS MAKES 2D VIEWS (makeView below), which a 1D, 1D-array or 3D image does
// not admit. A caller that fills viewType from the image gets a refusal here, and the
// wire arm's GenerateWireMipmap asks the same question before it grows the chain.
for (const VkImageViewType viewType : {source.viewType, destination.viewType}) {
if (viewType == VK_IMAGE_VIEW_TYPE_1D || viewType == VK_IMAGE_VIEW_TYPE_1D_ARRAY ||
viewType == VK_IMAGE_VIEW_TYPE_3D)
return false;
}
if (source.format != destination.format) return false;
// DEPTH ONLY. A combined depth/stencil image can be sampled for its depth and
// rendered as a depth attachment, but this pass writes gl_FragDepth and NOTHING
// else - the stencil aspect would come out of the render pass undefined, which is
// not what GL's glGenerateMipmap promises for the level it just produced. The
// caller declines that shape rather than producing half a level.
if (source.aspect != VK_IMAGE_ASPECT_DEPTH_BIT ||
destination.aspect != VK_IMAGE_ASPECT_DEPTH_BIT) return false;
if (!source.extent.width || !source.extent.height ||
!destination.extent.width || !destination.extent.height) return false;
VkFormatProperties properties{};
vkGetPhysicalDeviceFormatProperties(m_physicalDevice.handle, source.format, &properties);
const VkFormatFeatureFlags required = VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT |
VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT;
if ((properties.optimalTilingFeatures & required) != required) return false;
// The baked block's layout, and the offsets are not a guess: the generated header
// decorates its members at 0 / 16 / 32 / 40, which is where a vec4, a vec4, an int
// and an 8-byte-aligned ivec2 land. The pad is what makes the C++ agree.
struct Constants {
float srcRect[4];
float dstRect[4];
Int32 surfaceTransform;
Int32 reservedForIvec2Alignment;
Int32 srcTexelSize[2];
};
static_assert(sizeof(Constants) == 48);
static_assert(offsetof(Constants, srcTexelSize) == 40);
if (!m_wireDepthMipmapResources) {
m_wireDepthMipmapResources = new WireDepthMipmapResources;
auto& cache = *m_wireDepthMipmapResources;
const auto shader = [&](const Uint32* code, SizeT bytes, VkShaderModule& module) {
VkShaderModuleCreateInfo info{VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO};
info.codeSize = bytes;
info.pCode = code;
VK_VERIFY(vkCreateShaderModule(m_device, &info, nullptr, &module));
};
shader(kWireDepthMipmapVertexSpirv, sizeof(kWireDepthMipmapVertexSpirv), cache.vertex);
shader(kWireDepthMipmapFragmentSpirv, sizeof(kWireDepthMipmapFragmentSpirv), cache.fragment);
const VkDescriptorSetLayoutBinding binding{0, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
1, VK_SHADER_STAGE_FRAGMENT_BIT, nullptr};
VkDescriptorSetLayoutCreateInfo layout{VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO};
layout.bindingCount = 1;
layout.pBindings = &binding;
VK_VERIFY(vkCreateDescriptorSetLayout(m_device, &layout, nullptr, &cache.descriptorLayout));
// BOTH STAGES ON ONE RANGE: the block is declared identically in the vertex and
// fragment sources (a push-constant block must be), the vertex stage reads the
// rects and the fragment stage reads srcTexelSize.
const VkPushConstantRange push{VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT,
0, sizeof(Constants)};
VkPipelineLayoutCreateInfo pipelineLayout{VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO};
pipelineLayout.setLayoutCount = 1;
pipelineLayout.pSetLayouts = &cache.descriptorLayout;
pipelineLayout.pushConstantRangeCount = 1;
pipelineLayout.pPushConstantRanges = &push;
VK_VERIFY(vkCreatePipelineLayout(m_device, &pipelineLayout, nullptr, &cache.pipelineLayout));
// NEAREST, and it is not a default: the fragment shader does its own 2x2 box with
// texelFetch, so any filtering the sampler added would be applied twice. It also
// keeps the sampler legal on a device that does not advertise linear filtering for
// depth formats, which is most of them.
VkSamplerCreateInfo sampler{VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO};
sampler.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
sampler.addressModeU = sampler.addressModeV = sampler.addressModeW =
VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sampler.magFilter = sampler.minFilter = VK_FILTER_NEAREST;
sampler.maxLod = 0.0f;
VK_VERIFY(vkCreateSampler(m_device, &sampler, nullptr, &cache.nearest));
}
auto& cache = *m_wireDepthMipmapResources;
const VkImageLayout sourceLayout = sameImage ? VK_IMAGE_LAYOUT_GENERAL
: VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
const VkImageLayout destinationLayout = sameImage ? VK_IMAGE_LAYOUT_GENERAL
: VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
WireDepthMipmapResources::Target* target = nullptr;
for (auto& entry : cache.targets) {
if (entry.format == destination.format && entry.layout == destinationLayout) {
target = &entry;
break;
}
}
if (!target) {
cache.targets.emplace_back();
target = &cache.targets.back();
target->format = destination.format;
target->layout = destinationLayout;
VkAttachmentDescription attachment{};
attachment.format = destination.format;
attachment.samples = VK_SAMPLE_COUNT_1_BIT;
// DONT_CARE, not LOAD: the pass writes gl_FragDepth at every texel of the level
// it is producing, so there is nothing in the destination worth loading, and a
// level that has just been grown has no defined contents to load anyway.
attachment.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
attachment.initialLayout = attachment.finalLayout = destinationLayout;
const VkAttachmentReference depth{0, destinationLayout};
VkSubpassDescription subpass{};
subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass.pDepthStencilAttachment = &depth;
VkRenderPassCreateInfo pass{VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO};
pass.attachmentCount = 1;
pass.pAttachments = &attachment;
pass.subpassCount = 1;
pass.pSubpasses = &subpass;
VK_VERIFY(vkCreateRenderPass(m_device, &pass, nullptr, &target->renderPass));
VkPipelineShaderStageCreateInfo stages[2]{};
for (auto& stage : stages) {
stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stage.pName = "main";
}
stages[0].stage = VK_SHADER_STAGE_VERTEX_BIT;
stages[0].module = cache.vertex;
stages[1].stage = VK_SHADER_STAGE_FRAGMENT_BIT;
stages[1].module = cache.fragment;
const VkPipelineVertexInputStateCreateInfo vertex{VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO};
VkPipelineInputAssemblyStateCreateInfo assembly{VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO};
assembly.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
VkPipelineViewportStateCreateInfo viewport{VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO};
viewport.viewportCount = viewport.scissorCount = 1;
VkPipelineRasterizationStateCreateInfo raster{VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO};
raster.polygonMode = VK_POLYGON_MODE_FILL;
raster.cullMode = VK_CULL_MODE_NONE;
raster.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;
raster.lineWidth = 1.0f;
VkPipelineMultisampleStateCreateInfo samples{VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO};
samples.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;
// ALWAYS + write enabled: the fragment shader's gl_FragDepth is the answer, and a
// comparison against whatever DONT_CARE left behind would discard some of it.
VkPipelineDepthStencilStateCreateInfo depthStencil{VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO};
depthStencil.depthTestEnable = VK_TRUE;
depthStencil.depthWriteEnable = VK_TRUE;
depthStencil.depthCompareOp = VK_COMPARE_OP_ALWAYS;
depthStencil.minDepthBounds = 0.0f;
depthStencil.maxDepthBounds = 1.0f;
// No colour attachment in this render pass, so no attachment states either.
VkPipelineColorBlendStateCreateInfo blend{VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO};
const VkDynamicState dynamicStates[]{VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR};
VkPipelineDynamicStateCreateInfo dynamic{VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO};
dynamic.dynamicStateCount = 2;
dynamic.pDynamicStates = dynamicStates;
VkGraphicsPipelineCreateInfo pipeline{VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO};
pipeline.stageCount = 2;
pipeline.pStages = stages;
pipeline.pVertexInputState = &vertex;
pipeline.pInputAssemblyState = &assembly;
pipeline.pViewportState = &viewport;
pipeline.pRasterizationState = &raster;
pipeline.pMultisampleState = &samples;
pipeline.pDepthStencilState = &depthStencil;
pipeline.pColorBlendState = &blend;
pipeline.pDynamicState = &dynamic;
pipeline.layout = cache.pipelineLayout;
pipeline.renderPass = target->renderPass;
VK_VERIFY(vkCreateGraphicsPipelines(m_device, VK_NULL_HANDLE, 1, &pipeline, nullptr, &target->pipeline));
}
m_textureManager->FlushPendingUploads();
// Distinct mip views of the same image are legal shader input and output at once.
// GENERAL keeps the whole-image layout tracker truthful; the transition also orders
// the previous level's attachment write against this level's sample of it.
TransitionWireImage(source, sourceLayout);
if (sameImage) destination.layout = source.layout;
else TransitionWireImage(destination, destinationLayout);
const auto makeView = [&](const WireImage& image) {
VkImageViewCreateInfo view{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
view.image = image.image;
view.viewType = VK_IMAGE_VIEW_TYPE_2D;
view.format = image.format;
view.subresourceRange = {VK_IMAGE_ASPECT_DEPTH_BIT, image.level, 1, image.layer, 1};
VkImageView handle = VK_NULL_HANDLE;
VK_VERIFY(vkCreateImageView(m_device, &view, nullptr, &handle));
return handle;
};
const VkImageView sourceView = makeView(source), destinationView = makeView(destination);
VkFramebufferCreateInfo framebufferInfo{VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO};
framebufferInfo.renderPass = target->renderPass;
framebufferInfo.attachmentCount = 1;
framebufferInfo.pAttachments = &destinationView;
framebufferInfo.width = destination.extent.width;
framebufferInfo.height = destination.extent.height;
framebufferInfo.layers = 1;
VkFramebuffer framebuffer = VK_NULL_HANDLE;
VK_VERIFY(vkCreateFramebuffer(m_device, &framebufferInfo, nullptr, &framebuffer));
// Each in-flight pass owns its descriptor set, for the colour blit's reason: updating
// a cached set would rewrite the source view of an older queued pass.
VkDescriptorPool descriptorPool = VK_NULL_HANDLE;
VkDescriptorSet descriptorSet = VK_NULL_HANDLE;
const VkDescriptorPoolSize poolSize{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1};
VkDescriptorPoolCreateInfo pool{VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO};
pool.maxSets = 1;
pool.poolSizeCount = 1;
pool.pPoolSizes = &poolSize;
VK_VERIFY(vkCreateDescriptorPool(m_device, &pool, nullptr, &descriptorPool));
VkDescriptorSetAllocateInfo alloc{VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO};
alloc.descriptorPool = descriptorPool;
alloc.descriptorSetCount = 1;
alloc.pSetLayouts = &cache.descriptorLayout;
VK_VERIFY(vkAllocateDescriptorSets(m_device, &alloc, &descriptorSet));
const VkDescriptorImageInfo sampled{cache.nearest, sourceView, sourceLayout};
VkWriteDescriptorSet write{VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET};
write.dstSet = descriptorSet;
write.dstBinding = 0;
write.descriptorCount = 1;
write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
write.pImageInfo = &sampled;
vkUpdateDescriptorSets(m_device, 1, &write, 0, nullptr);
// The destination rectangle is flipped exactly as BlitWireColorImage flips it for an
// offscreen target, because it shares that file's vertex shader and therefore its
// window-Y flip. With the flip undone, destination row r samples texCoord.y
// (r + 0.5) / dstHeight, and the fragment's ivec2(texCoord * srcTexelSize) lands on
// the same source texels the monolith arm's fullscreen triangle lands on - which is
// the point: §3.2 retires a refusal by making the wire arm observably the monolith's,
// half-texel offset included.
const Constants constants{
{0.0f, 0.0f, 1.0f, 1.0f},
{0.0f, 1.0f, 1.0f, -1.0f},
0, 0,
{static_cast<Int32>(source.extent.width), static_cast<Int32>(source.extent.height)}};
auto& frame = m_frameContext.GetCurrent();
VkRenderPassBeginInfo begin{VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO};
begin.renderPass = target->renderPass;
begin.framebuffer = framebuffer;
begin.renderArea = {{0, 0}, destination.extent};
vkCmdBeginRenderPass(frame.commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, target->pipeline);
const VkViewport viewport{0.0f, 0.0f, static_cast<float>(destination.extent.width),
static_cast<float>(destination.extent.height), 0.0f, 1.0f};
const VkRect2D scissor{{0, 0}, destination.extent};
vkCmdSetViewport(frame.commandBuffer, 0, 1, &viewport);
vkCmdSetScissor(frame.commandBuffer, 0, 1, &scissor);
vkCmdBindDescriptorSets(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS,
cache.pipelineLayout, 0, 1, &descriptorSet, 0, nullptr);
vkCmdPushConstants(frame.commandBuffer, cache.pipelineLayout,
VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT,
0, sizeof(constants), &constants);
vkCmdDraw(frame.commandBuffer, 6, 1, 0, 0);
vkCmdEndRenderPass(frame.commandBuffer);
ResetDynamicStateShadow();
InvalidatePipelineMemo();
// This native pass binds its own set/layout without going through the application's
// descriptor binder. Its next fast path must rebind.
if (m_uniformManager) m_uniformManager->OnCommandBufferBoundary();
WireRetiredObjects retired{};
retired.framebuffers.push_back(framebuffer);
retired.imageViews = {sourceView, destinationView};
retired.descriptorPools.push_back(descriptorPool);
RetireWireObjects(Move(retired));
return true;
}
#endif
@@ -0,0 +1,33 @@
#version 450
// Native Vulkan depth mip generation: no GL program, registry entry, or runtime transpilation.
//
// The geometry is WireColorBlit.vert's, unchanged - the same quad, the same push-constant
// rect, the same GL-bottom-left to native-image mapping - so a depth mip and a colour blit
// place their fragments identically and only the fragment stage differs. The block carries
// one member more than the colour blit's, and the fragment stage is the member's only
// reader: a push-constant block must be declared identically in every stage of one pipeline,
// so it is declared here too rather than split into a second range.
layout(push_constant) uniform DepthMipmapRect {
vec4 srcRect;
vec4 dstRect;
int surfaceTransform;
ivec2 srcTexelSize;
} rect;
layout(location = 0) out vec2 texCoord;
void main() {
// A quad, rather than an oversized triangle, also clips partial blits to
// their destination rectangle. The Vulkan viewport covers the native image.
const vec2 corners[6] = vec2[](
vec2(0.0, 0.0), vec2(1.0, 0.0), vec2(0.0, 1.0),
vec2(0.0, 1.0), vec2(1.0, 0.0), vec2(1.0, 1.0));
vec2 uv = corners[gl_VertexIndex];
vec2 p = (rect.dstRect.xy + uv * rect.dstRect.zw) * 2.0 - 1.0;
// Same GL-bottom-left to native-image mapping as the monolithic blit.
p.y = -p.y;
if (rect.surfaceTransform == 1) p = vec2(-p.y, p.x);
else if (rect.surfaceTransform == 2) p = -p;
else if (rect.surfaceTransform == 3) p = vec2(p.y, -p.x);
gl_Position = vec4(p, 0.0, 1.0);
texCoord = rect.srcRect.xy + uv * rect.srcRect.zw;
}
@@ -0,0 +1,174 @@
// Generated from WireDepthMipmap.vert / WireDepthMipmap.frag by
// scripts/bake_internal_shaders.py, which runs MG_Test's BakedInternalShadersTest with the
// IN-TREE glslang and splices the words it emits into the arrays below. Do not edit the
// words by hand: MOBILEGL_BAKED_INTERNAL_SHADERS (the same test, run as a gate) recompiles
// every source in its table and reds on the first word that disagrees.
#pragma once
static constexpr Uint32 kWireDepthMipmapVertexSpirv[] = {
0x07230203u, 0x00010000u, 0x0008000bu, 0x00000071u, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
0x0008000fu, 0x00000000u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00000016u, 0x0000005eu, 0x00000067u,
0x00030003u, 0x00000002u, 0x000001c2u, 0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00030005u,
0x00000009u, 0x00007675u, 0x00060005u, 0x00000016u, 0x565f6c67u, 0x65747265u, 0x646e4978u, 0x00007865u,
0x00050005u, 0x00000019u, 0x65646e69u, 0x6c626178u, 0x00000065u, 0x00030005u, 0x0000001cu, 0x00000070u,
0x00060005u, 0x0000001fu, 0x74706544u, 0x70694d68u, 0x5270616du, 0x00746365u, 0x00050006u, 0x0000001fu,
0x00000000u, 0x52637273u, 0x00746365u, 0x00050006u, 0x0000001fu, 0x00000001u, 0x52747364u, 0x00746365u,
0x00080006u, 0x0000001fu, 0x00000002u, 0x66727573u, 0x54656361u, 0x736e6172u, 0x6d726f66u, 0x00000000u,
0x00070006u, 0x0000001fu, 0x00000003u, 0x54637273u, 0x6c657865u, 0x657a6953u, 0x00000000u, 0x00040005u,
0x00000021u, 0x74636572u, 0x00000000u, 0x00060005u, 0x0000005cu, 0x505f6c67u, 0x65567265u, 0x78657472u,
0x00000000u, 0x00060006u, 0x0000005cu, 0x00000000u, 0x505f6c67u, 0x7469736fu, 0x006e6f69u, 0x00070006u,
0x0000005cu, 0x00000001u, 0x505f6c67u, 0x746e696fu, 0x657a6953u, 0x00000000u, 0x00070006u, 0x0000005cu,
0x00000002u, 0x435f6c67u, 0x4470696cu, 0x61747369u, 0x0065636eu, 0x00070006u, 0x0000005cu, 0x00000003u,
0x435f6c67u, 0x446c6c75u, 0x61747369u, 0x0065636eu, 0x00030005u, 0x0000005eu, 0x00000000u, 0x00050005u,
0x00000067u, 0x43786574u, 0x64726f6fu, 0x00000000u, 0x00040047u, 0x00000016u, 0x0000000bu, 0x0000002au,
0x00030047u, 0x0000001fu, 0x00000002u, 0x00050048u, 0x0000001fu, 0x00000000u, 0x00000023u, 0x00000000u,
0x00050048u, 0x0000001fu, 0x00000001u, 0x00000023u, 0x00000010u, 0x00050048u, 0x0000001fu, 0x00000002u,
0x00000023u, 0x00000020u, 0x00050048u, 0x0000001fu, 0x00000003u, 0x00000023u, 0x00000028u, 0x00030047u,
0x0000005cu, 0x00000002u, 0x00050048u, 0x0000005cu, 0x00000000u, 0x0000000bu, 0x00000000u, 0x00050048u,
0x0000005cu, 0x00000001u, 0x0000000bu, 0x00000001u, 0x00050048u, 0x0000005cu, 0x00000002u, 0x0000000bu,
0x00000003u, 0x00050048u, 0x0000005cu, 0x00000003u, 0x0000000bu, 0x00000004u, 0x00040047u, 0x00000067u,
0x0000001eu, 0x00000000u, 0x00020013u, 0x00000002u, 0x00030021u, 0x00000003u, 0x00000002u, 0x00030016u,
0x00000006u, 0x00000020u, 0x00040017u, 0x00000007u, 0x00000006u, 0x00000002u, 0x00040020u, 0x00000008u,
0x00000007u, 0x00000007u, 0x00040015u, 0x0000000au, 0x00000020u, 0x00000000u, 0x0004002bu, 0x0000000au,
0x0000000bu, 0x00000006u, 0x0004001cu, 0x0000000cu, 0x00000007u, 0x0000000bu, 0x0004002bu, 0x00000006u,
0x0000000du, 0x00000000u, 0x0005002cu, 0x00000007u, 0x0000000eu, 0x0000000du, 0x0000000du, 0x0004002bu,
0x00000006u, 0x0000000fu, 0x3f800000u, 0x0005002cu, 0x00000007u, 0x00000010u, 0x0000000fu, 0x0000000du,
0x0005002cu, 0x00000007u, 0x00000011u, 0x0000000du, 0x0000000fu, 0x0005002cu, 0x00000007u, 0x00000012u,
0x0000000fu, 0x0000000fu, 0x0009002cu, 0x0000000cu, 0x00000013u, 0x0000000eu, 0x00000010u, 0x00000011u,
0x00000011u, 0x00000010u, 0x00000012u, 0x00040015u, 0x00000014u, 0x00000020u, 0x00000001u, 0x00040020u,
0x00000015u, 0x00000001u, 0x00000014u, 0x0004003bu, 0x00000015u, 0x00000016u, 0x00000001u, 0x00040020u,
0x00000018u, 0x00000007u, 0x0000000cu, 0x00040017u, 0x0000001du, 0x00000006u, 0x00000004u, 0x00040017u,
0x0000001eu, 0x00000014u, 0x00000002u, 0x0006001eu, 0x0000001fu, 0x0000001du, 0x0000001du, 0x00000014u,
0x0000001eu, 0x00040020u, 0x00000020u, 0x00000009u, 0x0000001fu, 0x0004003bu, 0x00000020u, 0x00000021u,
0x00000009u, 0x0004002bu, 0x00000014u, 0x00000022u, 0x00000001u, 0x00040020u, 0x00000023u, 0x00000009u,
0x0000001du, 0x0004002bu, 0x00000006u, 0x0000002du, 0x40000000u, 0x0004002bu, 0x0000000au, 0x00000031u,
0x00000001u, 0x00040020u, 0x00000032u, 0x00000007u, 0x00000006u, 0x0004002bu, 0x00000014u, 0x00000037u,
0x00000002u, 0x00040020u, 0x00000038u, 0x00000009u, 0x00000014u, 0x00020014u, 0x0000003bu, 0x0004002bu,
0x0000000au, 0x00000042u, 0x00000000u, 0x0004002bu, 0x00000014u, 0x00000051u, 0x00000003u, 0x0004001cu,
0x0000005bu, 0x00000006u, 0x00000031u, 0x0006001eu, 0x0000005cu, 0x0000001du, 0x00000006u, 0x0000005bu,
0x0000005bu, 0x00040020u, 0x0000005du, 0x00000003u, 0x0000005cu, 0x0004003bu, 0x0000005du, 0x0000005eu,
0x00000003u, 0x0004002bu, 0x00000014u, 0x0000005fu, 0x00000000u, 0x00040020u, 0x00000064u, 0x00000003u,
0x0000001du, 0x00040020u, 0x00000066u, 0x00000003u, 0x00000007u, 0x0004003bu, 0x00000066u, 0x00000067u,
0x00000003u, 0x00050036u, 0x00000002u, 0x00000004u, 0x00000000u, 0x00000003u, 0x000200f8u, 0x00000005u,
0x0004003bu, 0x00000008u, 0x00000009u, 0x00000007u, 0x0004003bu, 0x00000018u, 0x00000019u, 0x00000007u,
0x0004003bu, 0x00000008u, 0x0000001cu, 0x00000007u, 0x0004003du, 0x00000014u, 0x00000017u, 0x00000016u,
0x0003003eu, 0x00000019u, 0x00000013u, 0x00050041u, 0x00000008u, 0x0000001au, 0x00000019u, 0x00000017u,
0x0004003du, 0x00000007u, 0x0000001bu, 0x0000001au, 0x0003003eu, 0x00000009u, 0x0000001bu, 0x00050041u,
0x00000023u, 0x00000024u, 0x00000021u, 0x00000022u, 0x0004003du, 0x0000001du, 0x00000025u, 0x00000024u,
0x0007004fu, 0x00000007u, 0x00000026u, 0x00000025u, 0x00000025u, 0x00000000u, 0x00000001u, 0x0004003du,
0x00000007u, 0x00000027u, 0x00000009u, 0x00050041u, 0x00000023u, 0x00000028u, 0x00000021u, 0x00000022u,
0x0004003du, 0x0000001du, 0x00000029u, 0x00000028u, 0x0007004fu, 0x00000007u, 0x0000002au, 0x00000029u,
0x00000029u, 0x00000002u, 0x00000003u, 0x00050085u, 0x00000007u, 0x0000002bu, 0x00000027u, 0x0000002au,
0x00050081u, 0x00000007u, 0x0000002cu, 0x00000026u, 0x0000002bu, 0x0005008eu, 0x00000007u, 0x0000002eu,
0x0000002cu, 0x0000002du, 0x00050050u, 0x00000007u, 0x0000002fu, 0x0000000fu, 0x0000000fu, 0x00050083u,
0x00000007u, 0x00000030u, 0x0000002eu, 0x0000002fu, 0x0003003eu, 0x0000001cu, 0x00000030u, 0x00050041u,
0x00000032u, 0x00000033u, 0x0000001cu, 0x00000031u, 0x0004003du, 0x00000006u, 0x00000034u, 0x00000033u,
0x0004007fu, 0x00000006u, 0x00000035u, 0x00000034u, 0x00050041u, 0x00000032u, 0x00000036u, 0x0000001cu,
0x00000031u, 0x0003003eu, 0x00000036u, 0x00000035u, 0x00050041u, 0x00000038u, 0x00000039u, 0x00000021u,
0x00000037u, 0x0004003du, 0x00000014u, 0x0000003au, 0x00000039u, 0x000500aau, 0x0000003bu, 0x0000003cu,
0x0000003au, 0x00000022u, 0x000300f7u, 0x0000003eu, 0x00000000u, 0x000400fau, 0x0000003cu, 0x0000003du,
0x00000046u, 0x000200f8u, 0x0000003du, 0x00050041u, 0x00000032u, 0x0000003fu, 0x0000001cu, 0x00000031u,
0x0004003du, 0x00000006u, 0x00000040u, 0x0000003fu, 0x0004007fu, 0x00000006u, 0x00000041u, 0x00000040u,
0x00050041u, 0x00000032u, 0x00000043u, 0x0000001cu, 0x00000042u, 0x0004003du, 0x00000006u, 0x00000044u,
0x00000043u, 0x00050050u, 0x00000007u, 0x00000045u, 0x00000041u, 0x00000044u, 0x0003003eu, 0x0000001cu,
0x00000045u, 0x000200f9u, 0x0000003eu, 0x000200f8u, 0x00000046u, 0x00050041u, 0x00000038u, 0x00000047u,
0x00000021u, 0x00000037u, 0x0004003du, 0x00000014u, 0x00000048u, 0x00000047u, 0x000500aau, 0x0000003bu,
0x00000049u, 0x00000048u, 0x00000037u, 0x000300f7u, 0x0000004bu, 0x00000000u, 0x000400fau, 0x00000049u,
0x0000004au, 0x0000004eu, 0x000200f8u, 0x0000004au, 0x0004003du, 0x00000007u, 0x0000004cu, 0x0000001cu,
0x0004007fu, 0x00000007u, 0x0000004du, 0x0000004cu, 0x0003003eu, 0x0000001cu, 0x0000004du, 0x000200f9u,
0x0000004bu, 0x000200f8u, 0x0000004eu, 0x00050041u, 0x00000038u, 0x0000004fu, 0x00000021u, 0x00000037u,
0x0004003du, 0x00000014u, 0x00000050u, 0x0000004fu, 0x000500aau, 0x0000003bu, 0x00000052u, 0x00000050u,
0x00000051u, 0x000300f7u, 0x00000054u, 0x00000000u, 0x000400fau, 0x00000052u, 0x00000053u, 0x00000054u,
0x000200f8u, 0x00000053u, 0x00050041u, 0x00000032u, 0x00000055u, 0x0000001cu, 0x00000031u, 0x0004003du,
0x00000006u, 0x00000056u, 0x00000055u, 0x00050041u, 0x00000032u, 0x00000057u, 0x0000001cu, 0x00000042u,
0x0004003du, 0x00000006u, 0x00000058u, 0x00000057u, 0x0004007fu, 0x00000006u, 0x00000059u, 0x00000058u,
0x00050050u, 0x00000007u, 0x0000005au, 0x00000056u, 0x00000059u, 0x0003003eu, 0x0000001cu, 0x0000005au,
0x000200f9u, 0x00000054u, 0x000200f8u, 0x00000054u, 0x000200f9u, 0x0000004bu, 0x000200f8u, 0x0000004bu,
0x000200f9u, 0x0000003eu, 0x000200f8u, 0x0000003eu, 0x0004003du, 0x00000007u, 0x00000060u, 0x0000001cu,
0x00050051u, 0x00000006u, 0x00000061u, 0x00000060u, 0x00000000u, 0x00050051u, 0x00000006u, 0x00000062u,
0x00000060u, 0x00000001u, 0x00070050u, 0x0000001du, 0x00000063u, 0x00000061u, 0x00000062u, 0x0000000du,
0x0000000fu, 0x00050041u, 0x00000064u, 0x00000065u, 0x0000005eu, 0x0000005fu, 0x0003003eu, 0x00000065u,
0x00000063u, 0x00050041u, 0x00000023u, 0x00000068u, 0x00000021u, 0x0000005fu, 0x0004003du, 0x0000001du,
0x00000069u, 0x00000068u, 0x0007004fu, 0x00000007u, 0x0000006au, 0x00000069u, 0x00000069u, 0x00000000u,
0x00000001u, 0x0004003du, 0x00000007u, 0x0000006bu, 0x00000009u, 0x00050041u, 0x00000023u, 0x0000006cu,
0x00000021u, 0x0000005fu, 0x0004003du, 0x0000001du, 0x0000006du, 0x0000006cu, 0x0007004fu, 0x00000007u,
0x0000006eu, 0x0000006du, 0x0000006du, 0x00000002u, 0x00000003u, 0x00050085u, 0x00000007u, 0x0000006fu,
0x0000006bu, 0x0000006eu, 0x00050081u, 0x00000007u, 0x00000070u, 0x0000006au, 0x0000006fu, 0x0003003eu,
0x00000067u, 0x00000070u, 0x000100fdu, 0x00010038u,
};
static constexpr Uint32 kWireDepthMipmapFragmentSpirv[] = {
0x07230203u, 0x00010000u, 0x0008000bu, 0x00000059u, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
0x0007000fu, 0x00000004u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x0000000du, 0x0000004fu, 0x00030010u,
0x00000004u, 0x00000007u, 0x00030010u, 0x00000004u, 0x0000000cu, 0x00030003u, 0x00000002u, 0x000001c2u,
0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00040005u, 0x00000009u, 0x42637273u, 0x00657361u,
0x00050005u, 0x0000000du, 0x43786574u, 0x64726f6fu, 0x00000000u, 0x00060005u, 0x00000010u, 0x74706544u,
0x70694d68u, 0x5270616du, 0x00746365u, 0x00050006u, 0x00000010u, 0x00000000u, 0x52637273u, 0x00746365u,
0x00050006u, 0x00000010u, 0x00000001u, 0x52747364u, 0x00746365u, 0x00080006u, 0x00000010u, 0x00000002u,
0x66727573u, 0x54656361u, 0x736e6172u, 0x6d726f66u, 0x00000000u, 0x00070006u, 0x00000010u, 0x00000003u,
0x54637273u, 0x6c657865u, 0x657a6953u, 0x00000000u, 0x00040005u, 0x00000012u, 0x74636572u, 0x00000000u,
0x00040005u, 0x0000001au, 0x4d637273u, 0x00007861u, 0x00040005u, 0x00000021u, 0x74706564u, 0x00003068u,
0x00050005u, 0x00000025u, 0x72756f73u, 0x6d496563u, 0x00656761u, 0x00040005u, 0x00000031u, 0x74706564u,
0x00003168u, 0x00040005u, 0x0000003bu, 0x74706564u, 0x00003268u, 0x00040005u, 0x00000045u, 0x74706564u,
0x00003368u, 0x00060005u, 0x0000004fu, 0x465f6c67u, 0x44676172u, 0x68747065u, 0x00000000u, 0x00040047u,
0x0000000du, 0x0000001eu, 0x00000000u, 0x00030047u, 0x00000010u, 0x00000002u, 0x00050048u, 0x00000010u,
0x00000000u, 0x00000023u, 0x00000000u, 0x00050048u, 0x00000010u, 0x00000001u, 0x00000023u, 0x00000010u,
0x00050048u, 0x00000010u, 0x00000002u, 0x00000023u, 0x00000020u, 0x00050048u, 0x00000010u, 0x00000003u,
0x00000023u, 0x00000028u, 0x00040047u, 0x00000025u, 0x00000021u, 0x00000000u, 0x00040047u, 0x00000025u,
0x00000022u, 0x00000000u, 0x00040047u, 0x0000004fu, 0x0000000bu, 0x00000016u, 0x00020013u, 0x00000002u,
0x00030021u, 0x00000003u, 0x00000002u, 0x00040015u, 0x00000006u, 0x00000020u, 0x00000001u, 0x00040017u,
0x00000007u, 0x00000006u, 0x00000002u, 0x00040020u, 0x00000008u, 0x00000007u, 0x00000007u, 0x00030016u,
0x0000000au, 0x00000020u, 0x00040017u, 0x0000000bu, 0x0000000au, 0x00000002u, 0x00040020u, 0x0000000cu,
0x00000001u, 0x0000000bu, 0x0004003bu, 0x0000000cu, 0x0000000du, 0x00000001u, 0x00040017u, 0x0000000fu,
0x0000000au, 0x00000004u, 0x0006001eu, 0x00000010u, 0x0000000fu, 0x0000000fu, 0x00000006u, 0x00000007u,
0x00040020u, 0x00000011u, 0x00000009u, 0x00000010u, 0x0004003bu, 0x00000011u, 0x00000012u, 0x00000009u,
0x0004002bu, 0x00000006u, 0x00000013u, 0x00000003u, 0x00040020u, 0x00000014u, 0x00000009u, 0x00000007u,
0x0004002bu, 0x00000006u, 0x0000001du, 0x00000001u, 0x0005002cu, 0x00000007u, 0x0000001eu, 0x0000001du,
0x0000001du, 0x00040020u, 0x00000020u, 0x00000007u, 0x0000000au, 0x00090019u, 0x00000022u, 0x0000000au,
0x00000001u, 0x00000000u, 0x00000000u, 0x00000000u, 0x00000001u, 0x00000000u, 0x0003001bu, 0x00000023u,
0x00000022u, 0x00040020u, 0x00000024u, 0x00000000u, 0x00000023u, 0x0004003bu, 0x00000024u, 0x00000025u,
0x00000000u, 0x0004002bu, 0x00000006u, 0x00000028u, 0x00000000u, 0x0005002cu, 0x00000007u, 0x00000029u,
0x00000028u, 0x00000028u, 0x00040015u, 0x0000002eu, 0x00000020u, 0x00000000u, 0x0004002bu, 0x0000002eu,
0x0000002fu, 0x00000000u, 0x0005002cu, 0x00000007u, 0x00000034u, 0x0000001du, 0x00000028u, 0x0005002cu,
0x00000007u, 0x0000003eu, 0x00000028u, 0x0000001du, 0x00040020u, 0x0000004eu, 0x00000003u, 0x0000000au,
0x0004003bu, 0x0000004eu, 0x0000004fu, 0x00000003u, 0x0004002bu, 0x0000000au, 0x00000050u, 0x3e800000u,
0x00050036u, 0x00000002u, 0x00000004u, 0x00000000u, 0x00000003u, 0x000200f8u, 0x00000005u, 0x0004003bu,
0x00000008u, 0x00000009u, 0x00000007u, 0x0004003bu, 0x00000008u, 0x0000001au, 0x00000007u, 0x0004003bu,
0x00000020u, 0x00000021u, 0x00000007u, 0x0004003bu, 0x00000020u, 0x00000031u, 0x00000007u, 0x0004003bu,
0x00000020u, 0x0000003bu, 0x00000007u, 0x0004003bu, 0x00000020u, 0x00000045u, 0x00000007u, 0x0004003du,
0x0000000bu, 0x0000000eu, 0x0000000du, 0x00050041u, 0x00000014u, 0x00000015u, 0x00000012u, 0x00000013u,
0x0004003du, 0x00000007u, 0x00000016u, 0x00000015u, 0x0004006fu, 0x0000000bu, 0x00000017u, 0x00000016u,
0x00050085u, 0x0000000bu, 0x00000018u, 0x0000000eu, 0x00000017u, 0x0004006eu, 0x00000007u, 0x00000019u,
0x00000018u, 0x0003003eu, 0x00000009u, 0x00000019u, 0x00050041u, 0x00000014u, 0x0000001bu, 0x00000012u,
0x00000013u, 0x0004003du, 0x00000007u, 0x0000001cu, 0x0000001bu, 0x00050082u, 0x00000007u, 0x0000001fu,
0x0000001cu, 0x0000001eu, 0x0003003eu, 0x0000001au, 0x0000001fu, 0x0004003du, 0x00000023u, 0x00000026u,
0x00000025u, 0x0004003du, 0x00000007u, 0x00000027u, 0x00000009u, 0x0004003du, 0x00000007u, 0x0000002au,
0x0000001au, 0x0008000cu, 0x00000007u, 0x0000002bu, 0x00000001u, 0x0000002du, 0x00000027u, 0x00000029u,
0x0000002au, 0x00040064u, 0x00000022u, 0x0000002cu, 0x00000026u, 0x0007005fu, 0x0000000fu, 0x0000002du,
0x0000002cu, 0x0000002bu, 0x00000002u, 0x00000028u, 0x00050051u, 0x0000000au, 0x00000030u, 0x0000002du,
0x00000000u, 0x0003003eu, 0x00000021u, 0x00000030u, 0x0004003du, 0x00000023u, 0x00000032u, 0x00000025u,
0x0004003du, 0x00000007u, 0x00000033u, 0x00000009u, 0x00050080u, 0x00000007u, 0x00000035u, 0x00000033u,
0x00000034u, 0x0004003du, 0x00000007u, 0x00000036u, 0x0000001au, 0x0008000cu, 0x00000007u, 0x00000037u,
0x00000001u, 0x0000002du, 0x00000035u, 0x00000029u, 0x00000036u, 0x00040064u, 0x00000022u, 0x00000038u,
0x00000032u, 0x0007005fu, 0x0000000fu, 0x00000039u, 0x00000038u, 0x00000037u, 0x00000002u, 0x00000028u,
0x00050051u, 0x0000000au, 0x0000003au, 0x00000039u, 0x00000000u, 0x0003003eu, 0x00000031u, 0x0000003au,
0x0004003du, 0x00000023u, 0x0000003cu, 0x00000025u, 0x0004003du, 0x00000007u, 0x0000003du, 0x00000009u,
0x00050080u, 0x00000007u, 0x0000003fu, 0x0000003du, 0x0000003eu, 0x0004003du, 0x00000007u, 0x00000040u,
0x0000001au, 0x0008000cu, 0x00000007u, 0x00000041u, 0x00000001u, 0x0000002du, 0x0000003fu, 0x00000029u,
0x00000040u, 0x00040064u, 0x00000022u, 0x00000042u, 0x0000003cu, 0x0007005fu, 0x0000000fu, 0x00000043u,
0x00000042u, 0x00000041u, 0x00000002u, 0x00000028u, 0x00050051u, 0x0000000au, 0x00000044u, 0x00000043u,
0x00000000u, 0x0003003eu, 0x0000003bu, 0x00000044u, 0x0004003du, 0x00000023u, 0x00000046u, 0x00000025u,
0x0004003du, 0x00000007u, 0x00000047u, 0x00000009u, 0x00050080u, 0x00000007u, 0x00000048u, 0x00000047u,
0x0000001eu, 0x0004003du, 0x00000007u, 0x00000049u, 0x0000001au, 0x0008000cu, 0x00000007u, 0x0000004au,
0x00000001u, 0x0000002du, 0x00000048u, 0x00000029u, 0x00000049u, 0x00040064u, 0x00000022u, 0x0000004bu,
0x00000046u, 0x0007005fu, 0x0000000fu, 0x0000004cu, 0x0000004bu, 0x0000004au, 0x00000002u, 0x00000028u,
0x00050051u, 0x0000000au, 0x0000004du, 0x0000004cu, 0x00000000u, 0x0003003eu, 0x00000045u, 0x0000004du,
0x0004003du, 0x0000000au, 0x00000051u, 0x00000021u, 0x0004003du, 0x0000000au, 0x00000052u, 0x00000031u,
0x00050081u, 0x0000000au, 0x00000053u, 0x00000051u, 0x00000052u, 0x0004003du, 0x0000000au, 0x00000054u,
0x0000003bu, 0x00050081u, 0x0000000au, 0x00000055u, 0x00000053u, 0x00000054u, 0x0004003du, 0x0000000au,
0x00000056u, 0x00000045u, 0x00050081u, 0x0000000au, 0x00000057u, 0x00000055u, 0x00000056u, 0x00050085u,
0x0000000au, 0x00000058u, 0x00000050u, 0x00000057u, 0x0003003eu, 0x0000004fu, 0x00000058u, 0x000100fdu,
0x00010038u,
};
@@ -0,0 +1,329 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WireDepthResolveArm.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Includes.h>
#include <cstring>
#include <mutex>
#include <utility>
// P7 gate 5 (g5-msrbo, g5-msprobe): WHICH ARM RESOLVES A MULTISAMPLE DEPTH/STENCIL ASPECT FIRST.
//
// The wire arm has two (WireFramebuffer.inc, ResolveWireDepthStencil): a render pass that carries
// a VK_KHR_depth_stencil_resolve attachment and no draw, and the baked shader pass of
// WireMultisampleResolve.inc, which draws. P7 wave 2-B2 took the render pass wherever the
// extension exists. On the Redmi (Adreno 830, Vulkan 1.3.284, driver 512.800.71) that pass left
// its resolve target unwritten - KHR-GL46.direct_state_access.renderbuffers_storage_multisample
// read every resolved depth and stencil value back as 0 on the inproc arm while the monolith arm
// passed, and the same case passed on the device with the shader arm forced
// (MGITEST_MAGMA_FORCE_SHADER_DEPTH_RESOLVE=1).
//
// A DRIVER THAT CLAIMS WHAT IT CANNOT DO IS A POST PROBE, NOT A VENDOR QUIRK (standing rule,
// 2026-08-22). The first fix keyed the order on Qualcomm's vendor id; it is replaced by a probe
// that reproduces the production resolve on the actual VkDevice (WireDepthResolveProbe.cpp) and
// asks the one question the order depends on: does the no-draw render pass write the resolve
// target? A driver that fixes the defect gets the fast arm back, and a driver of any vendor that
// has it gets the shader arm - neither needs a table edit.
//
// EVERY BUG PROBE CARRIES A CONTROL (MG_Util/SelfTest/DriverBugProbes.h). The control is the
// production SHADER arm fed the same inputs: when it does not resolve them either, the probe has
// measured its own setup (or a device that cannot resolve depth at all) rather than the render
// pass, the verdict is INCONCLUSIVE, and the default order stands.
//
// This header is the pure half - the measurement, the verdict, the test knob and the choice - so
// a unit test pins every mapping with a fake result source (MG_Test/Pipeline/PipelineQuirkTest.cpp).
// The Vulkan half that fills the measurement is WireDepthResolveProbe.{h,cpp}.
namespace MobileGL::MG_Backend::DirectVulkan {
// One aspect of one probed format, both arms. Values are the RAW aspect words the
// buffer copy returns (D16: 16 bits; D24: the low 24 bits of the 32-bit word; D32F: the float's
// bits; stencil: 8 bits), compared within `tolerance` (1 LSB for a UNORM depth, 0 otherwise).
struct WireDepthResolveAspectReading {
// The format has this aspect and the render-pass arm's copy of it was read back.
Bool measured = false;
Uint32 texels = 0;
Uint32 expected = 0; // the multisample source's cleared value
Uint32 sentinel = 0; // what the resolve target was pre-filled with
Uint32 tolerance = 0;
// THE SUBJECT: VK_KHR_depth_stencil_resolve render pass, no draw.
Uint32 renderPassMatches = 0; // texels within tolerance of `expected`
Uint32 renderPassSentinels = 0; // texels still holding the sentinel
Uint32 renderPassFirst = 0; // texel (0,0)
// THE CONTROL: the production shader pass, same inputs. False where it cannot run for this
// aspect (a stencil aspect without VK_EXT_shader_stencil_export, a failed pipeline).
Bool shaderRan = false;
Uint32 shaderMatches = 0;
Uint32 shaderFirst = 0;
};
struct WireDepthResolveFormatReading {
String name; // e.g. "D24_UNORM_S8_UINT"
Int32 format = 0; // the VkFormat
Uint32 samples = 0;
// False when the format was not probed; `skipReason` says why (unsupported, the render-pass
// arm would not be taken for it, an object the probe needed could not be created).
Bool ran = false;
String skipReason;
WireDepthResolveAspectReading depth, stencil;
};
struct WireDepthResolveProbeMeasurement {
// At least one format was recorded, submitted and read back.
Bool ran = false;
// The bounded fence wait expired: every object the probe created was deliberately leaked
// (the queue may still be executing it), and nothing was read back.
Bool fenceWaitTimedOut = false;
String failureReason;
// Came from MGITEST_MAGMA_DEPTH_RESOLVE_PROBE=bug|clean, not from the device.
Bool fromKnob = false;
// Measured on the device, but with MGITEST_MAGMA_DEPTH_RESOLVE_PROBE=elide-subject: the probe
// recorded no render-pass resolve at all, so the subject's target kept its sentinel.
Bool subjectElided = false;
Vector<WireDepthResolveFormatReading> formats;
};
enum class WireDepthResolveProbeVerdict : Uint8 {
// Nothing was measured: the render-pass arm is not available (nothing to choose between),
// or the probe could not set itself up. The default order stands.
NotRun,
// Measured, but no format's CONTROL resolved: the probe cannot tell the render pass's
// defect from its own. The default order stands, and the report must never call it a bug.
Inconclusive,
// Every format whose control resolved was resolved by the render pass too.
Clean,
// The defect: on at least one format the shader control resolved the inputs and the
// render pass did not (wrote nothing - the sentinel survived - or wrote wrong values).
RenderPassResolveBroken,
};
inline const char* WireDepthResolveProbeVerdictName(WireDepthResolveProbeVerdict verdict) {
switch (verdict) {
case WireDepthResolveProbeVerdict::NotRun: return "not-run";
case WireDepthResolveProbeVerdict::Inconclusive: return "inconclusive";
case WireDepthResolveProbeVerdict::Clean: return "clean";
case WireDepthResolveProbeVerdict::RenderPassResolveBroken: return "render-pass-resolve-broken";
}
return "?";
}
inline Bool WireDepthResolveControlHolds(const WireDepthResolveAspectReading& aspect) {
return aspect.measured && aspect.shaderRan && aspect.texels > 0 && aspect.shaderMatches == aspect.texels;
}
inline Bool WireDepthResolveSubjectHolds(const WireDepthResolveAspectReading& aspect) {
return aspect.texels > 0 && aspect.renderPassMatches == aspect.texels;
}
// An aspect only counts where its control holds; a format is conclusive when any of its
// aspects is. The render pass resolves every aspect of the format in one pass, so one aspect
// it fails is the defect for the whole arm.
inline WireDepthResolveProbeVerdict EvaluateWireDepthResolveProbe(const WireDepthResolveProbeMeasurement& measurement) {
if (!measurement.ran || measurement.fenceWaitTimedOut) return WireDepthResolveProbeVerdict::NotRun;
Bool conclusive = false, broken = false;
for (const WireDepthResolveFormatReading& reading : measurement.formats) {
if (!reading.ran) continue;
for (const WireDepthResolveAspectReading* aspect : {&reading.depth, &reading.stencil}) {
if (!WireDepthResolveControlHolds(*aspect)) continue;
conclusive = true;
if (!WireDepthResolveSubjectHolds(*aspect)) broken = true;
}
}
if (broken) return WireDepthResolveProbeVerdict::RenderPassResolveBroken;
return conclusive ? WireDepthResolveProbeVerdict::Clean : WireDepthResolveProbeVerdict::Inconclusive;
}
// THE TEST KNOB, read by the SERVER (split and spawn arms only; the tcp server is a lane fixture
// whose environment no entry reaches - spawn_lane_parity.py's MAGMA_SERVER_ENV_KNOB_NO_TCP).
// Lavapipe resolves correctly through both arms, so no host lane can reach the verdict that
// matters on the Redmi. MGITEST_MAGMA_DEPTH_RESOLVE_PROBE=bug|clean replaces the MEASUREMENT
// with a canned one that says exactly that, and nothing else: the canned measurement still goes
// through EvaluateWireDepthResolveProbe and the same member, so an entry that forces `bug` is red
// when the evaluation stops detecting the defect.
//
// A canned measurement cannot see the REAL probe - its recording, readback and tally - stop
// detecting the defect, and on lavapipe the real subject is clean, so nothing would. The third
// value, MGITEST_MAGMA_DEPTH_RESOLVE_PROBE=elide-subject, is that negative control: the REAL
// probe runs (ChooseWireDepthResolveArm treats it as Measure) and the caller passes it on as
// WireDepthResolveProbeContext::elideSubject, so the probe records no render-pass resolve and
// the real readback keeps the sentinel while the shader control still runs. The
// DirectVulkan.{Split,Spawn}.MsResolveElide. entries then require the defect verdict from the
// measurement itself.
enum class WireDepthResolveProbeKnob : Uint8 { Measure, ForceBug, ForceClean, ElideSubject, Unrecognised };
inline WireDepthResolveProbeKnob ParseWireDepthResolveProbeKnob(const char* value) {
if (value == nullptr || *value == '\0') return WireDepthResolveProbeKnob::Measure;
if (std::strcmp(value, "bug") == 0) return WireDepthResolveProbeKnob::ForceBug;
if (std::strcmp(value, "clean") == 0) return WireDepthResolveProbeKnob::ForceClean;
if (std::strcmp(value, "elide-subject") == 0) return WireDepthResolveProbeKnob::ElideSubject;
return WireDepthResolveProbeKnob::Unrecognised;
}
// One packed format's worth of readings: the control resolved both aspects; the render pass
// either did too (`renderPassWrites`) or left the sentinel everywhere.
inline WireDepthResolveProbeMeasurement CannedWireDepthResolveProbeMeasurement(Bool renderPassWrites) {
WireDepthResolveProbeMeasurement measurement;
measurement.ran = true;
measurement.fromKnob = true;
WireDepthResolveFormatReading reading;
reading.name = "MGITEST_MAGMA_DEPTH_RESOLVE_PROBE";
reading.samples = 4;
reading.ran = true;
const auto fill = [&](WireDepthResolveAspectReading& aspect, Uint32 expected, Uint32 sentinel) {
aspect.measured = true;
aspect.texels = 16;
aspect.expected = expected;
aspect.sentinel = sentinel;
aspect.renderPassMatches = renderPassWrites ? 16 : 0;
aspect.renderPassSentinels = renderPassWrites ? 0 : 16;
aspect.renderPassFirst = renderPassWrites ? expected : sentinel;
aspect.shaderRan = true;
aspect.shaderMatches = 16;
aspect.shaderFirst = expected;
};
fill(reading.depth, 0x400000u, 0xBFFFFFu);
fill(reading.stencil, 0x5Au, 0xA5u);
measurement.formats.push_back(Move(reading));
return measurement;
}
struct WireDepthResolveArmChoice {
WireDepthResolveProbeMeasurement measurement;
WireDepthResolveProbeVerdict verdict = WireDepthResolveProbeVerdict::NotRun;
// m_wirePreferShaderDepthResolve: the shader pass goes first, the render pass is the
// fallback for what the shader cannot write.
Bool preferShader = false;
};
// The whole decision, with the Vulkan half behind `measure` (called at most once, and only when
// the knob does not decide and the render-pass arm exists at all: without it the shader pass
// is the only arm and there is no order to choose). ElideSubject does not decide: it is
// measured like Measure, and the elision is the measuring callable's to apply.
template <typename MeasureFn>
WireDepthResolveArmChoice ChooseWireDepthResolveArm(WireDepthResolveProbeKnob knob, Bool renderPassArmAvailable,
MeasureFn&& measure) {
WireDepthResolveArmChoice choice;
if (knob == WireDepthResolveProbeKnob::ForceBug || knob == WireDepthResolveProbeKnob::ForceClean) {
choice.measurement = CannedWireDepthResolveProbeMeasurement(knob == WireDepthResolveProbeKnob::ForceClean);
} else if (renderPassArmAvailable) {
choice.measurement = measure();
}
choice.verdict = EvaluateWireDepthResolveProbe(choice.measurement);
choice.preferShader = choice.verdict == WireDepthResolveProbeVerdict::RenderPassResolveBroken;
return choice;
}
// WHOSE VERDICT IT IS (codex closeout finding 7). The verdict is a property of ONE device and
// driver, and a process can hold more than one: inproc tears a renderer down and initializes
// another, possibly on a different physical device or under a different driver. The choice was
// a function-static - the first renderer's answer for every renderer after it - so a second
// device skipped its own probe and took an order measured elsewhere. It is now memoized per
// IDENTITY: the device (vendor, device, driver version, pipeline-cache UUID - the last changes
// with the driver build even where the version number does not) plus the two inputs the choice
// itself depends on (whether a render-pass arm exists to judge, and the test knob). The same
// identity is still probed once per process; a different one is probed for itself.
struct WireDepthResolveDeviceIdentity {
Uint32 vendorID = 0;
Uint32 deviceID = 0;
Uint32 driverVersion = 0;
Uint8 pipelineCacheUUID[16] = {};
Bool renderPassArmAvailable = false;
WireDepthResolveProbeKnob knob = WireDepthResolveProbeKnob::Measure;
Bool operator==(const WireDepthResolveDeviceIdentity& other) const {
return vendorID == other.vendorID && deviceID == other.deviceID && driverVersion == other.driverVersion &&
std::memcmp(pipelineCacheUUID, other.pipelineCacheUUID, sizeof(pipelineCacheUUID)) == 0 &&
renderPassArmAvailable == other.renderPassArmAvailable && knob == other.knob;
}
};
// The renderer's identity from its device's VkPhysicalDeviceProperties - the ONE place the fields
// are filled (VulkanRenderer::ArmWireDepthResolveOrder calls it), so the unit test proves that each
// property reaches the key: with only the cache tested, a field left out here stayed green.
inline WireDepthResolveDeviceIdentity MakeWireDepthResolveDeviceIdentity(const VkPhysicalDeviceProperties& properties,
Bool renderPassArmAvailable,
WireDepthResolveProbeKnob knob) {
WireDepthResolveDeviceIdentity identity;
identity.vendorID = properties.vendorID;
identity.deviceID = properties.deviceID;
identity.driverVersion = properties.driverVersion;
static_assert(sizeof(identity.pipelineCacheUUID) == sizeof(properties.pipelineCacheUUID));
std::memcpy(identity.pipelineCacheUUID, properties.pipelineCacheUUID, sizeof(identity.pipelineCacheUUID));
identity.renderPassArmAvailable = renderPassArmAvailable;
identity.knob = knob;
return identity;
}
class WireDepthResolveArmCache {
public:
// The identity's memoized choice, or `decide()`'s - called at most once per identity, under
// the cache's lock (a probe records on the device's queue; two renderers of one identity
// must not both run it). `decided`, when given, says whether this call ran `decide`.
template <typename DecideFn>
WireDepthResolveArmChoice Resolve(const WireDepthResolveDeviceIdentity& identity, DecideFn&& decide,
Bool* decided = nullptr) {
std::lock_guard<std::mutex> lock(m_mutex);
for (const auto& entry : m_entries) {
if (entry.first == identity) {
if (decided) *decided = false;
return entry.second;
}
}
WireDepthResolveArmChoice choice = decide();
m_entries.emplace_back(identity, choice);
if (decided) *decided = true;
return choice;
}
SizeT Size() const {
std::lock_guard<std::mutex> lock(m_mutex);
return m_entries.size();
}
private:
mutable std::mutex m_mutex;
Vector<std::pair<WireDepthResolveDeviceIdentity, WireDepthResolveArmChoice>> m_entries;
};
// One format's readings, both arms: "D24_UNORM_S8_UINT x4: depth expected 0x400000 sentinel
// 0xbfffff: render pass 0/16 (first 0xbfffff, sentinel 16), shader control 16/16 (first
// 0x400000) / stencil ...".
inline String DescribeWireDepthResolveFormat(const WireDepthResolveFormatReading& reading) {
const auto aspectText = [](const char* name, const WireDepthResolveAspectReading& aspect) {
String text = format("{} expected 0x{:x} sentinel 0x{:x}: render pass {}/{} (first 0x{:x}, sentinel {}), ",
name, aspect.expected, aspect.sentinel, aspect.renderPassMatches, aspect.texels,
aspect.renderPassFirst, aspect.renderPassSentinels);
if (aspect.shaderRan) {
text += format("shader control {}/{} (first 0x{:x})", aspect.shaderMatches, aspect.texels,
aspect.shaderFirst);
} else {
text += "no shader control";
}
return text;
};
String out = reading.name;
if (!reading.ran) return out + " skipped (" + reading.skipReason + ")";
out += format(" x{}: ", reading.samples);
if (reading.depth.measured) out += aspectText("depth", reading.depth);
if (reading.depth.measured && reading.stencil.measured) out += " / ";
if (reading.stencil.measured) out += aspectText("stencil", reading.stencil);
return out;
}
// Every format, "; "-separated, for the POST row.
inline String DescribeWireDepthResolveProbe(const WireDepthResolveProbeMeasurement& measurement) {
if (!measurement.ran) {
return measurement.failureReason.empty() ? String("not run") : "not run (" + measurement.failureReason + ")";
}
String out;
for (const WireDepthResolveFormatReading& reading : measurement.formats) {
if (!out.empty()) out += "; ";
out += DescribeWireDepthResolveFormat(reading);
}
return out;
}
} // namespace MobileGL::MG_Backend::DirectVulkan
@@ -0,0 +1,992 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WireDepthResolveProbe.cpp
// Copyright (c) 2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
//
// See WireDepthResolveProbe.h for the arms and the probe's shape, WireDepthResolveArm.h for the
// verdict. The plumbing is the PRIMITIVES_GENERATED probe's (MG_Util/SelfTest/
// PrimitivesGeneratedNoXfbProbe.cpp): one throwaway command buffer, a bounded fence wait that
// leaks rather than idle-waits a hung GPU, teardown on every other path.
#include "WireDepthResolveProbe.h"
#include <MG_Util/Debug/Log.h>
#include <bit>
#include <cmath>
#include <cstring>
#include <vulkan/vulkan_core.h>
#if MOBILEGL_BUILD_DISAGGREGATED
namespace MobileGL::MG_Backend::DirectVulkan {
#include "WireMultisampleResolveSpirv.h"
VkImageAspectFlags WireDepthStencilFormatAspects(VkFormat format) {
if (format == VK_FORMAT_S8_UINT) return VK_IMAGE_ASPECT_STENCIL_BIT;
if (format == VK_FORMAT_D16_UNORM_S8_UINT || format == VK_FORMAT_D24_UNORM_S8_UINT ||
format == VK_FORMAT_D32_SFLOAT_S8_UINT)
return VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
return VK_IMAGE_ASPECT_DEPTH_BIT;
}
VkImageViewCreateInfo WireDepthStencilViewInfo(VkImage image, VkFormat format, VkImageAspectFlags aspects,
Uint32 level, Uint32 layer) {
VkImageViewCreateInfo info{VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO};
info.image = image;
info.viewType = VK_IMAGE_VIEW_TYPE_2D;
info.format = format;
info.subresourceRange = {aspects, level, 1, layer, 1};
return info;
}
VkResult CreateWireDepthStencilResolvePass(PFN_vkCreateRenderPass2 createRenderPass2, VkDevice device,
VkFormat format, VkSampleCountFlagBits samples, VkRenderPass* pass) {
const VkImageAspectFlags formatAspects = WireDepthStencilFormatAspects(format);
VkAttachmentDescription2 attachments[2]{};
for (Uint32 i = 0; i < 2; ++i) {
auto& attachment = attachments[i];
attachment.sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2;
attachment.format = format;
attachment.samples = i == 0 ? samples : VK_SAMPLE_COUNT_1_BIT;
attachment.loadOp = attachment.stencilLoadOp =
i == 0 ? VK_ATTACHMENT_LOAD_OP_LOAD : VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachment.storeOp = attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
attachment.initialLayout = attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
}
VkAttachmentReference2 sourceReference{VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2};
sourceReference.attachment = 0;
sourceReference.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
sourceReference.aspectMask = formatAspects;
VkAttachmentReference2 resolveReference = sourceReference;
resolveReference.attachment = 1;
VkSubpassDescriptionDepthStencilResolve resolve{VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE};
resolve.depthResolveMode =
(formatAspects & VK_IMAGE_ASPECT_DEPTH_BIT) ? VK_RESOLVE_MODE_SAMPLE_ZERO_BIT : VK_RESOLVE_MODE_NONE;
resolve.stencilResolveMode =
(formatAspects & VK_IMAGE_ASPECT_STENCIL_BIT) ? VK_RESOLVE_MODE_SAMPLE_ZERO_BIT : VK_RESOLVE_MODE_NONE;
resolve.pDepthStencilResolveAttachment = &resolveReference;
VkSubpassDescription2 subpass{VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2};
subpass.pNext = &resolve;
subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass.pDepthStencilAttachment = &sourceReference;
VkRenderPassCreateInfo2 passInfo{VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2};
passInfo.attachmentCount = 2;
passInfo.pAttachments = attachments;
passInfo.subpassCount = 1;
passInfo.pSubpasses = &subpass;
return createRenderPass2(device, &passInfo, nullptr, pass);
}
VkResult CreateWireShaderResolveKit(VkDevice device, Bool shaderStencilExport, WireShaderResolveKit& kit) {
const auto shader = [&](const Uint32* code, SizeT bytes, VkShaderModule& module) {
VkShaderModuleCreateInfo info{VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO};
info.codeSize = bytes;
info.pCode = code;
return vkCreateShaderModule(device, &info, nullptr, &module);
};
VkResult result =
shader(kWireMultisampleResolveVertexSpirv, sizeof(kWireMultisampleResolveVertexSpirv), kit.vertex);
if (result != VK_SUCCESS) return result;
result = shader(kWireMultisampleDepthResolveFragmentSpirv, sizeof(kWireMultisampleDepthResolveFragmentSpirv),
kit.depthFragment);
if (result != VK_SUCCESS) return result;
// THE STENCIL MODULE IS ONLY CREATED WHERE IT CAN RUN: its SPIR-V declares the
// StencilExportEXT capability, and handing a module with an unsupported capability to
// vkCreateShaderModule is undefined behaviour - not an error this could catch.
if (shaderStencilExport) {
result = shader(kWireMultisampleStencilResolveFragmentSpirv,
sizeof(kWireMultisampleStencilResolveFragmentSpirv), kit.stencilFragment);
if (result != VK_SUCCESS) return result;
}
const VkDescriptorSetLayoutBinding binding{0, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1,
VK_SHADER_STAGE_FRAGMENT_BIT, nullptr};
VkDescriptorSetLayoutCreateInfo layout{VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO};
layout.bindingCount = 1;
layout.pBindings = &binding;
result = vkCreateDescriptorSetLayout(device, &layout, nullptr, &kit.descriptorLayout);
if (result != VK_SUCCESS) return result;
// NO PUSH CONSTANT RANGE AT ALL: the pass reads its coordinate from gl_FragCoord and its
// rectangle from the scissor, so neither stage takes an input.
VkPipelineLayoutCreateInfo pipelineLayout{VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO};
pipelineLayout.setLayoutCount = 1;
pipelineLayout.pSetLayouts = &kit.descriptorLayout;
result = vkCreatePipelineLayout(device, &pipelineLayout, nullptr, &kit.pipelineLayout);
if (result != VK_SUCCESS) return result;
// NEAREST: texelFetch ignores the sampler's filter, and most devices do not advertise
// linear filtering for a depth format anyway. The sampler exists only because GLSL's
// sampler2DMS is a COMBINED image sampler.
VkSamplerCreateInfo sampler{VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO};
sampler.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
sampler.addressModeU = sampler.addressModeV = sampler.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sampler.magFilter = sampler.minFilter = VK_FILTER_NEAREST;
sampler.maxLod = 0.0f;
return vkCreateSampler(device, &sampler, nullptr, &kit.nearest);
}
void DestroyWireShaderResolveKit(VkDevice device, WireShaderResolveKit& kit) {
vkDestroySampler(device, kit.nearest, nullptr);
vkDestroyPipelineLayout(device, kit.pipelineLayout, nullptr);
vkDestroyDescriptorSetLayout(device, kit.descriptorLayout, nullptr);
vkDestroyShaderModule(device, kit.vertex, nullptr);
vkDestroyShaderModule(device, kit.depthFragment, nullptr);
vkDestroyShaderModule(device, kit.stencilFragment, nullptr);
kit = {};
}
VkResult CreateWireShaderResolveRenderPass(VkDevice device, VkFormat format, VkRenderPass* pass) {
VkAttachmentDescription attachment{};
attachment.format = format;
attachment.samples = VK_SAMPLE_COUNT_1_BIT;
// DONT_CARE on BOTH aspects: the pass writes every texel of the scissored rect for the
// aspect it owns, and the caller copies only that aspect out. The other aspect of a packed
// format is left undefined on purpose - it is never read, and LOADing it would order this
// pass behind a write nobody is waiting for.
attachment.loadOp = attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachment.storeOp = attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
attachment.initialLayout = attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
const VkAttachmentReference depthStencil{0, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL};
VkSubpassDescription subpass{};
subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass.pDepthStencilAttachment = &depthStencil;
VkRenderPassCreateInfo info{VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO};
info.attachmentCount = 1;
info.pAttachments = &attachment;
info.subpassCount = 1;
info.pSubpasses = &subpass;
return vkCreateRenderPass(device, &info, nullptr, pass);
}
VkResult CreateWireShaderResolvePipeline(VkDevice device, const WireShaderResolveKit& kit, VkRenderPass pass,
VkImageAspectFlags aspect, VkPipeline* pipeline) {
const VkShaderModule fragment =
aspect == VK_IMAGE_ASPECT_STENCIL_BIT ? kit.stencilFragment : kit.depthFragment;
if (fragment == VK_NULL_HANDLE || kit.vertex == VK_NULL_HANDLE) return VK_ERROR_FEATURE_NOT_PRESENT;
VkPipelineShaderStageCreateInfo stages[2]{};
for (auto& stage : stages) {
stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stage.pName = "main";
}
stages[0].stage = VK_SHADER_STAGE_VERTEX_BIT;
stages[0].module = kit.vertex;
stages[1].stage = VK_SHADER_STAGE_FRAGMENT_BIT;
stages[1].module = fragment;
const VkPipelineVertexInputStateCreateInfo vertex{VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO};
VkPipelineInputAssemblyStateCreateInfo assembly{VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO};
assembly.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
VkPipelineViewportStateCreateInfo viewport{VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO};
viewport.viewportCount = viewport.scissorCount = 1;
VkPipelineRasterizationStateCreateInfo raster{VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO};
raster.polygonMode = VK_POLYGON_MODE_FILL;
raster.cullMode = VK_CULL_MODE_NONE;
raster.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;
raster.lineWidth = 1.0f;
VkPipelineMultisampleStateCreateInfo samples{VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO};
samples.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;
VkPipelineDepthStencilStateCreateInfo depthStencilState{VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO};
depthStencilState.minDepthBounds = 0.0f;
depthStencilState.maxDepthBounds = 1.0f;
if (aspect == VK_IMAGE_ASPECT_DEPTH_BIT) {
// ALWAYS + write: gl_FragDepth is the answer, and a comparison against whatever
// DONT_CARE left behind would discard some of it.
depthStencilState.depthTestEnable = VK_TRUE;
depthStencilState.depthWriteEnable = VK_TRUE;
depthStencilState.depthCompareOp = VK_COMPARE_OP_ALWAYS;
} else {
// REPLACE on every outcome, compare ALWAYS, full write mask: that is what makes the
// shader-exported reference the value that lands. The static `reference` is overridden
// by gl_FragStencilRefARB, which is the whole point of the extension.
depthStencilState.stencilTestEnable = VK_TRUE;
VkStencilOpState op{};
op.failOp = op.passOp = op.depthFailOp = VK_STENCIL_OP_REPLACE;
op.compareOp = VK_COMPARE_OP_ALWAYS;
op.compareMask = op.writeMask = 0xFFu;
op.reference = 0;
depthStencilState.front = depthStencilState.back = op;
}
// No colour attachment in this render pass, so no attachment states either.
VkPipelineColorBlendStateCreateInfo blend{VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO};
const VkDynamicState dynamicStates[]{VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR};
VkPipelineDynamicStateCreateInfo dynamic{VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO};
dynamic.dynamicStateCount = 2;
dynamic.pDynamicStates = dynamicStates;
VkGraphicsPipelineCreateInfo info{VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO};
info.stageCount = 2;
info.pStages = stages;
info.pVertexInputState = &vertex;
info.pInputAssemblyState = &assembly;
info.pViewportState = &viewport;
info.pRasterizationState = &raster;
info.pMultisampleState = &samples;
info.pDepthStencilState = &depthStencilState;
info.pColorBlendState = &blend;
info.pDynamicState = &dynamic;
info.layout = kit.pipelineLayout;
info.renderPass = pass;
return vkCreateGraphicsPipelines(device, VK_NULL_HANDLE, 1, &info, nullptr, pipeline);
}
void RecordWireShaderResolveDraw(VkCommandBuffer commandBuffer, const WireShaderResolveKit& kit,
VkRenderPass pass, VkPipeline pipeline, VkFramebuffer framebuffer,
VkDescriptorSet descriptorSet, VkExtent2D attachmentExtent, VkRect2D rect) {
VkRenderPassBeginInfo begin{VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO};
begin.renderPass = pass;
begin.framebuffer = framebuffer;
// The RECTANGLE is the render area, exactly as the render-pass arm's renderArea is - the
// scratch outside it is never read.
begin.renderArea = rect;
vkCmdBeginRenderPass(commandBuffer, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdBindPipeline(commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
// THE VIEWPORT IS THE WHOLE ATTACHMENT AND THE SCISSOR IS THE RECT, and the two are not
// interchangeable here: gl_FragCoord is what the fragment shader fetches by, so the
// clip-space-to-attachment mapping has to be the identity for every rectangle. A viewport
// shrunk onto the rect would slide the whole image instead of moving a window over it.
const VkViewport viewport{0.0f, 0.0f, static_cast<float>(attachmentExtent.width),
static_cast<float>(attachmentExtent.height), 0.0f, 1.0f};
vkCmdSetViewport(commandBuffer, 0, 1, &viewport);
vkCmdSetScissor(commandBuffer, 0, 1, &rect);
vkCmdBindDescriptorSets(commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, kit.pipelineLayout, 0, 1,
&descriptorSet, 0, nullptr);
vkCmdDraw(commandBuffer, 3, 1, 0, 0);
vkCmdEndRenderPass(commandBuffer);
}
namespace {
constexpr Uint32 kProbeExtent = 4;
constexpr Uint32 kProbeTexels = kProbeExtent * kProbeExtent;
// Every Vulkan device must support 4 samples on a depth/stencil attachment
// (framebufferDepthSampleCounts), and the CTS case the probe stands for fails at 2 and 4.
constexpr VkSampleCountFlagBits kProbeSamples = VK_SAMPLE_COUNT_4_BIT;
constexpr Float kClearDepth = 0.25f;
constexpr Float kSentinelDepth = 0.75f;
constexpr Uint32 kClearStencil = 0x5Au;
constexpr Uint32 kSentinelStencil = 0xA5u;
constexpr Uint64 kFenceTimeoutNs = 5ull * 1000ull * 1000ull * 1000ull;
// The formats the wire maps GL's depth/stencil internal formats onto (VkTextureManager),
// packed first: the case that set all this off failed on every one of them.
struct ProbeFormat {
VkFormat format;
const char* name;
};
constexpr ProbeFormat kProbeFormats[] = {
{VK_FORMAT_D24_UNORM_S8_UINT, "D24_UNORM_S8_UINT"},
{VK_FORMAT_D32_SFLOAT_S8_UINT, "D32_SFLOAT_S8_UINT"},
{VK_FORMAT_D16_UNORM, "D16_UNORM"},
{VK_FORMAT_X8_D24_UNORM_PACK32, "X8_D24_UNORM_PACK32"},
{VK_FORMAT_D32_SFLOAT, "D32_SFLOAT"},
{VK_FORMAT_S8_UINT, "S8_UINT"},
};
// The usages production gives the two images: a wire multisample depth/stencil attachment
// (VkTextureManager: SAMPLED | DEPTH_STENCIL_ATTACHMENT | TRANSFER_SRC | TRANSFER_DST) and
// ResolveWireDepthStencil's resolve scratch (DEPTH_STENCIL_ATTACHMENT | TRANSFER_SRC).
constexpr VkImageUsageFlags kSourceUsage = VK_IMAGE_USAGE_SAMPLED_BIT |
VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT |
VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;
constexpr VkImageUsageFlags kScratchUsage =
VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
Bool IsFloatDepth(VkFormat format) {
return format == VK_FORMAT_D32_SFLOAT || format == VK_FORMAT_D32_SFLOAT_S8_UINT;
}
Uint32 DepthCopyBytes(VkFormat format) { return format == VK_FORMAT_D16_UNORM ? 2u : 4u; }
Uint32 EncodeDepth(VkFormat format, Float depth) {
if (IsFloatDepth(format)) return std::bit_cast<Uint32>(depth);
const double scale = format == VK_FORMAT_D16_UNORM ? 65535.0 : 16777215.0;
return static_cast<Uint32>(std::lround(static_cast<double>(depth) * scale));
}
Uint32 ReadDepthWord(VkFormat format, const Uint8* texel) {
if (format == VK_FORMAT_D16_UNORM) {
Uint16 word = 0;
std::memcpy(&word, texel, sizeof(word));
return word;
}
Uint32 word = 0;
std::memcpy(&word, texel, sizeof(word));
// A D24 aspect copies out as a 32-bit word whose top byte is undefined.
return IsFloatDepth(format) ? word : (word & 0x00FFFFFFu);
}
Uint32 WordDistance(Uint32 a, Uint32 b) { return a > b ? a - b : b - a; }
struct ProbeImage {
VkImage image = VK_NULL_HANDLE;
VkDeviceMemory memory = VK_NULL_HANDLE;
};
struct ControlPass {
Bool ready = false;
ProbeImage target;
VkRenderPass pass = VK_NULL_HANDLE;
VkPipeline pipeline = VK_NULL_HANDLE;
VkImageView sampledView = VK_NULL_HANDLE;
VkImageView targetView = VK_NULL_HANDLE;
VkFramebuffer framebuffer = VK_NULL_HANDLE;
VkDescriptorSet set = VK_NULL_HANDLE;
VkDeviceSize offset = 0;
};
struct FormatWork {
SizeT reading = 0;
VkFormat format = VK_FORMAT_UNDEFINED;
VkImageAspectFlags aspects = 0;
ProbeImage subjectSource, controlSource, subjectTarget;
VkRenderPass resolvePass = VK_NULL_HANDLE;
VkImageView resolveViews[2]{};
VkFramebuffer resolveFramebuffer = VK_NULL_HANDLE;
VkDeviceSize subjectDepthOffset = 0, subjectStencilOffset = 0;
ControlPass depthControl, stencilControl;
};
// Everything the probe creates, destroyed together (or leaked together on a timeout).
struct ProbeObjects {
VkDevice device = VK_NULL_HANDLE;
Vector<VkImage> images;
Vector<VkDeviceMemory> memories;
Vector<VkImageView> views;
Vector<VkRenderPass> passes;
Vector<VkFramebuffer> framebuffers;
Vector<VkPipeline> pipelines;
VkBuffer buffer = VK_NULL_HANDLE;
VkDescriptorPool descriptorPool = VK_NULL_HANDLE;
VkCommandPool commandPool = VK_NULL_HANDLE;
VkFence fence = VK_NULL_HANDLE;
WireShaderResolveKit kit;
void Destroy() {
for (VkPipeline pipeline : pipelines) vkDestroyPipeline(device, pipeline, nullptr);
for (VkFramebuffer framebuffer : framebuffers) vkDestroyFramebuffer(device, framebuffer, nullptr);
for (VkRenderPass pass : passes) vkDestroyRenderPass(device, pass, nullptr);
for (VkImageView view : views) vkDestroyImageView(device, view, nullptr);
for (VkImage image : images) vkDestroyImage(device, image, nullptr);
vkDestroyBuffer(device, buffer, nullptr);
for (VkDeviceMemory memory : memories) vkFreeMemory(device, memory, nullptr);
vkDestroyDescriptorPool(device, descriptorPool, nullptr);
vkDestroyCommandPool(device, commandPool, nullptr);
vkDestroyFence(device, fence, nullptr);
DestroyWireShaderResolveKit(device, kit);
}
};
Optional<Uint32> PickMemoryType(const VkPhysicalDeviceMemoryProperties& properties, Uint32 allowed,
VkMemoryPropertyFlags wanted, VkMemoryPropertyFlags required) {
Optional<Uint32> fallback;
for (Uint32 i = 0; i < properties.memoryTypeCount; ++i) {
if (!(allowed & (1u << i))) continue;
const VkMemoryPropertyFlags flags = properties.memoryTypes[i].propertyFlags;
if (flags & VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT) continue;
if ((flags & required) != required) continue;
if ((flags & wanted) == wanted) return i;
if (!fallback) fallback = i;
}
return fallback;
}
void ImageBarrier(VkCommandBuffer commandBuffer, VkImage image, VkImageAspectFlags aspects,
VkImageLayout oldLayout, VkImageLayout newLayout, VkPipelineStageFlags srcStage,
VkAccessFlags srcAccess, VkPipelineStageFlags dstStage, VkAccessFlags dstAccess) {
VkImageMemoryBarrier barrier{VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER};
barrier.srcAccessMask = srcAccess;
barrier.dstAccessMask = dstAccess;
barrier.oldLayout = oldLayout;
barrier.newLayout = newLayout;
barrier.srcQueueFamilyIndex = barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.image = image;
barrier.subresourceRange = {aspects, 0, 1, 0, VK_REMAINING_ARRAY_LAYERS};
vkCmdPipelineBarrier(commandBuffer, srcStage, dstStage, 0, 0, nullptr, 0, nullptr, 1, &barrier);
}
} // namespace
WireDepthResolveProbeMeasurement RunWireDepthResolveProbe(const WireDepthResolveProbeContext& context) {
WireDepthResolveProbeMeasurement measurement;
measurement.subjectElided = context.elideSubject;
const VkDevice device = context.device;
if (device == VK_NULL_HANDLE || context.queue == VK_NULL_HANDLE || context.createRenderPass2 == nullptr) {
measurement.failureReason = "no device, queue or vkCreateRenderPass2";
return measurement;
}
ProbeObjects objects;
objects.device = device;
Bool leak = false;
struct Teardown {
ProbeObjects& objects;
Bool& leak;
~Teardown() {
if (!leak) objects.Destroy();
}
} teardown{objects, leak};
const auto fail = [&](String reason) {
measurement.ran = false;
measurement.failureReason = Move(reason);
return measurement;
};
VkPhysicalDeviceMemoryProperties memoryProperties{};
vkGetPhysicalDeviceMemoryProperties(context.physicalDevice, &memoryProperties);
const auto makeImage = [&](VkFormat format, VkSampleCountFlagBits samples, VkImageUsageFlags usage,
ProbeImage& out) -> Bool {
VkImageCreateInfo info{VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO};
info.imageType = VK_IMAGE_TYPE_2D;
info.format = format;
info.extent = {kProbeExtent, kProbeExtent, 1};
info.mipLevels = info.arrayLayers = 1;
info.samples = samples;
info.tiling = VK_IMAGE_TILING_OPTIMAL;
info.usage = usage;
info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
info.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
if (vkCreateImage(device, &info, nullptr, &out.image) != VK_SUCCESS) return false;
objects.images.push_back(out.image);
VkMemoryRequirements requirements{};
vkGetImageMemoryRequirements(device, out.image, &requirements);
const Optional<Uint32> type = PickMemoryType(memoryProperties, requirements.memoryTypeBits,
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, 0);
if (!type) return false;
VkMemoryAllocateInfo allocate{VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO};
allocate.allocationSize = requirements.size;
allocate.memoryTypeIndex = *type;
if (vkAllocateMemory(device, &allocate, nullptr, &out.memory) != VK_SUCCESS) return false;
objects.memories.push_back(out.memory);
return vkBindImageMemory(device, out.image, out.memory, 0) == VK_SUCCESS;
};
const auto makeView = [&](const VkImageViewCreateInfo& info, VkImageView& out) -> Bool {
if (vkCreateImageView(device, &info, nullptr, &out) != VK_SUCCESS) return false;
objects.views.push_back(out);
return true;
};
const auto makeFramebuffer = [&](VkRenderPass pass, const VkImageView* views, Uint32 count,
VkFramebuffer& out) -> Bool {
VkFramebufferCreateInfo info{VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO};
info.renderPass = pass;
info.attachmentCount = count;
info.pAttachments = views;
info.width = info.height = kProbeExtent;
info.layers = 1;
if (vkCreateFramebuffer(device, &info, nullptr, &out) != VK_SUCCESS) return false;
objects.framebuffers.push_back(out);
return true;
};
// The wire's clear (WireFramebuffer.inc, ClearWireFramebuffer): a LOAD/STORE pass over every
// aspect of the format with vkCmdClearAttachments inside it. Recorded later; built now.
struct ClearPass {
VkRenderPass pass = VK_NULL_HANDLE;
VkFramebuffer framebuffer = VK_NULL_HANDLE;
VkImageAspectFlags aspects = 0;
Float depth = 0.0f;
Uint32 stencil = 0;
};
Vector<ClearPass> clears;
const auto makeClear = [&](const ProbeImage& image, VkFormat format, VkSampleCountFlagBits samples,
Float depth, Uint32 stencil) -> Bool {
const VkImageAspectFlags aspects = WireDepthStencilFormatAspects(format);
VkImageView view = VK_NULL_HANDLE;
if (!makeView(WireDepthStencilViewInfo(image.image, format, aspects, 0, 0), view)) return false;
VkAttachmentDescription attachment{};
attachment.format = format;
attachment.samples = samples;
attachment.loadOp = attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
attachment.storeOp = attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
attachment.initialLayout = attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
const VkAttachmentReference reference{0, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL};
VkSubpassDescription subpass{};
subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass.pDepthStencilAttachment = &reference;
VkRenderPassCreateInfo info{VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO};
info.attachmentCount = 1;
info.pAttachments = &attachment;
info.subpassCount = 1;
info.pSubpasses = &subpass;
ClearPass clear;
if (vkCreateRenderPass(device, &info, nullptr, &clear.pass) != VK_SUCCESS) return false;
objects.passes.push_back(clear.pass);
if (!makeFramebuffer(clear.pass, &view, 1, clear.framebuffer)) return false;
clear.aspects = aspects;
clear.depth = depth;
clear.stencil = stencil;
clears.push_back(clear);
return true;
};
// THE CONTROL'S FORMAT-INDEPENDENT HALF. A kit that cannot be made leaves every control
// un-run, which the verdict reads as INCONCLUSIVE - never as the bug.
String kitFailure;
if (const VkResult result = CreateWireShaderResolveKit(device, context.shaderStencilExport, objects.kit);
result != VK_SUCCESS)
kitFailure = format("shader resolve kit creation failed (VkResult {})", static_cast<Int>(result));
// Readback layout: every region on a 16-byte boundary (a depth/stencil buffer copy wants a
// multiple of 4).
VkDeviceSize bufferBytes = 0;
const auto reserve = [&](VkDeviceSize bytes) {
const VkDeviceSize offset = bufferBytes;
bufferBytes += (bytes + 15u) & ~VkDeviceSize{15u};
return offset;
};
Vector<FormatWork> work;
for (const ProbeFormat& probeFormat : kProbeFormats) {
WireDepthResolveFormatReading reading;
reading.name = probeFormat.name;
reading.format = static_cast<Int32>(probeFormat.format);
reading.samples = static_cast<Uint32>(kProbeSamples);
const SizeT readingIndex = measurement.formats.size();
measurement.formats.push_back(reading);
auto& slot = measurement.formats.back();
// A skipped format records nothing: the clears it had already built are dropped with it
// (their images stay in `objects` and are only destroyed).
const SizeT clearsBefore = clears.size();
const auto skip = [&](String reason) {
slot.skipReason = Move(reason);
clears.resize(clearsBefore);
};
const VkFormat format = probeFormat.format;
const VkImageAspectFlags aspects = WireDepthStencilFormatAspects(format);
VkFormatProperties formatProperties{};
vkGetPhysicalDeviceFormatProperties(context.physicalDevice, format, &formatProperties);
constexpr VkFormatFeatureFlags kRequired =
VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT | VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT;
if ((formatProperties.optimalTilingFeatures & kRequired) != kRequired) {
skip("not a sampled depth/stencil attachment format here");
continue;
}
VkImageFormatProperties imageProperties{};
if (vkGetPhysicalDeviceImageFormatProperties(context.physicalDevice, format, VK_IMAGE_TYPE_2D,
VK_IMAGE_TILING_OPTIMAL, kSourceUsage, 0,
&imageProperties) != VK_SUCCESS ||
!(imageProperties.sampleCounts & kProbeSamples)) {
skip("no 4x multisample image with the wire's attachment usage");
continue;
}
if (vkGetPhysicalDeviceImageFormatProperties(context.physicalDevice, format, VK_IMAGE_TYPE_2D,
VK_IMAGE_TILING_OPTIMAL, kScratchUsage, 0,
&imageProperties) != VK_SUCCESS) {
skip("no single-sample image with the resolve scratch's usage");
continue;
}
// ResolveWireDepthStencil's own test for the render-pass arm, per format.
if (((aspects & VK_IMAGE_ASPECT_DEPTH_BIT) &&
!(context.depthResolveModes & VK_RESOLVE_MODE_SAMPLE_ZERO_BIT)) ||
((aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
!(context.stencilResolveModes & VK_RESOLVE_MODE_SAMPLE_ZERO_BIT))) {
skip("the render-pass arm is not taken for this format (no SAMPLE_ZERO resolve mode)");
continue;
}
FormatWork item;
item.reading = readingIndex;
item.format = format;
item.aspects = aspects;
if (!makeImage(format, kProbeSamples, kSourceUsage, item.subjectSource) ||
!makeImage(format, kProbeSamples, kSourceUsage, item.controlSource) ||
!makeImage(format, VK_SAMPLE_COUNT_1_BIT, kScratchUsage, item.subjectTarget)) {
skip("image creation failed");
continue;
}
if (CreateWireDepthStencilResolvePass(context.createRenderPass2, device, format, kProbeSamples,
&item.resolvePass) != VK_SUCCESS) {
skip("the render-pass arm's vkCreateRenderPass2 failed");
continue;
}
objects.passes.push_back(item.resolvePass);
if (!makeView(WireDepthStencilViewInfo(item.subjectSource.image, format, aspects, 0, 0),
item.resolveViews[0]) ||
!makeView(WireDepthStencilViewInfo(item.subjectTarget.image, format, aspects, 0, 0),
item.resolveViews[1]) ||
!makeFramebuffer(item.resolvePass, item.resolveViews, 2, item.resolveFramebuffer)) {
skip("the render-pass arm's views or framebuffer could not be created");
continue;
}
if (!makeClear(item.subjectSource, format, kProbeSamples, kClearDepth, kClearStencil) ||
!makeClear(item.controlSource, format, kProbeSamples, kClearDepth, kClearStencil) ||
!makeClear(item.subjectTarget, format, VK_SAMPLE_COUNT_1_BIT, kSentinelDepth, kSentinelStencil)) {
skip("the clear passes could not be created");
continue;
}
// The control, one target per aspect: the shader pass's attachment is DONT_CARE on both
// aspects, so a second aspect's pass into the same target would discard the first.
const auto makeControl = [&](VkImageAspectFlags aspect, ControlPass& control) {
if (!kitFailure.empty()) return;
if (aspect == VK_IMAGE_ASPECT_STENCIL_BIT && objects.kit.stencilFragment == VK_NULL_HANDLE) return;
if (!makeImage(format, VK_SAMPLE_COUNT_1_BIT, kScratchUsage, control.target)) return;
if (CreateWireShaderResolveRenderPass(device, format, &control.pass) != VK_SUCCESS) return;
objects.passes.push_back(control.pass);
if (CreateWireShaderResolvePipeline(device, objects.kit, control.pass, aspect, &control.pipeline) !=
VK_SUCCESS)
return;
objects.pipelines.push_back(control.pipeline);
if (!makeView(WireDepthStencilViewInfo(item.controlSource.image, format, aspect, 0, 0),
control.sampledView) ||
!makeView(WireDepthStencilViewInfo(control.target.image, format, aspects, 0, 0),
control.targetView) ||
!makeFramebuffer(control.pass, &control.targetView, 1, control.framebuffer))
return;
if (!makeClear(control.target, format, VK_SAMPLE_COUNT_1_BIT, kSentinelDepth, kSentinelStencil))
return;
control.ready = true;
};
if (aspects & VK_IMAGE_ASPECT_DEPTH_BIT) makeControl(VK_IMAGE_ASPECT_DEPTH_BIT, item.depthControl);
if (aspects & VK_IMAGE_ASPECT_STENCIL_BIT) makeControl(VK_IMAGE_ASPECT_STENCIL_BIT, item.stencilControl);
if (aspects & VK_IMAGE_ASPECT_DEPTH_BIT) {
item.subjectDepthOffset = reserve(kProbeTexels * DepthCopyBytes(format));
if (item.depthControl.ready) item.depthControl.offset = reserve(kProbeTexels * DepthCopyBytes(format));
}
if (aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
item.subjectStencilOffset = reserve(kProbeTexels);
if (item.stencilControl.ready) item.stencilControl.offset = reserve(kProbeTexels);
}
slot.ran = true;
work.push_back(item);
}
if (work.empty()) return fail("no format could be probed");
// Descriptor sets for the controls, one per draw (the production shader arm's own rule).
Uint32 controlDraws = 0;
for (const FormatWork& item : work)
controlDraws += (item.depthControl.ready ? 1u : 0u) + (item.stencilControl.ready ? 1u : 0u);
if (controlDraws > 0) {
const VkDescriptorPoolSize poolSize{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, controlDraws};
VkDescriptorPoolCreateInfo pool{VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO};
pool.maxSets = controlDraws;
pool.poolSizeCount = 1;
pool.pPoolSizes = &poolSize;
if (vkCreateDescriptorPool(device, &pool, nullptr, &objects.descriptorPool) != VK_SUCCESS)
return fail("vkCreateDescriptorPool failed");
for (FormatWork& item : work) {
for (ControlPass* control : {&item.depthControl, &item.stencilControl}) {
if (!control->ready) continue;
VkDescriptorSetAllocateInfo alloc{VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO};
alloc.descriptorPool = objects.descriptorPool;
alloc.descriptorSetCount = 1;
alloc.pSetLayouts = &objects.kit.descriptorLayout;
if (vkAllocateDescriptorSets(device, &alloc, &control->set) != VK_SUCCESS)
return fail("vkAllocateDescriptorSets failed");
const VkDescriptorImageInfo sampled{objects.kit.nearest, control->sampledView,
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL};
VkWriteDescriptorSet write{VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET};
write.dstSet = control->set;
write.dstBinding = 0;
write.descriptorCount = 1;
write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
write.pImageInfo = &sampled;
vkUpdateDescriptorSets(device, 1, &write, 0, nullptr);
}
}
}
// The host-visible readback buffer.
VkDeviceMemory bufferMemory = VK_NULL_HANDLE;
Bool bufferCoherent = false;
{
VkBufferCreateInfo info{VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO};
info.size = bufferBytes;
info.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT;
info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
if (vkCreateBuffer(device, &info, nullptr, &objects.buffer) != VK_SUCCESS)
return fail("vkCreateBuffer failed");
VkMemoryRequirements requirements{};
vkGetBufferMemoryRequirements(device, objects.buffer, &requirements);
const Optional<Uint32> type =
PickMemoryType(memoryProperties, requirements.memoryTypeBits, VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT);
if (!type) return fail("no host-visible memory type for the readback buffer");
bufferCoherent = (memoryProperties.memoryTypes[*type].propertyFlags &
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
VkMemoryAllocateInfo allocate{VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO};
allocate.allocationSize = requirements.size;
allocate.memoryTypeIndex = *type;
if (vkAllocateMemory(device, &allocate, nullptr, &bufferMemory) != VK_SUCCESS)
return fail("vkAllocateMemory (readback) failed");
objects.memories.push_back(bufferMemory);
if (vkBindBufferMemory(device, objects.buffer, bufferMemory, 0) != VK_SUCCESS)
return fail("vkBindBufferMemory failed");
}
VkCommandPoolCreateInfo poolInfo{VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO};
poolInfo.flags = VK_COMMAND_POOL_CREATE_TRANSIENT_BIT;
poolInfo.queueFamilyIndex = context.queueFamilyIndex;
if (vkCreateCommandPool(device, &poolInfo, nullptr, &objects.commandPool) != VK_SUCCESS)
return fail("vkCreateCommandPool failed");
VkCommandBufferAllocateInfo commandInfo{VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO};
commandInfo.commandPool = objects.commandPool;
commandInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
commandInfo.commandBufferCount = 1;
VkCommandBuffer cmd = VK_NULL_HANDLE;
if (vkAllocateCommandBuffers(device, &commandInfo, &cmd) != VK_SUCCESS)
return fail("vkAllocateCommandBuffers failed");
VkCommandBufferBeginInfo beginInfo{VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO};
beginInfo.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
if (vkBeginCommandBuffer(cmd, &beginInfo) != VK_SUCCESS) return fail("vkBeginCommandBuffer failed");
const VkRect2D fullRect{{0, 0}, {kProbeExtent, kProbeExtent}};
// (1) Every image UNDEFINED -> DEPTH_STENCIL_ATTACHMENT_OPTIMAL, the way TransitionWireImage
// moves a fresh image (source state from UNDEFINED, destination ALL_COMMANDS).
for (const FormatWork& item : work) {
const VkImage images[] = {item.subjectSource.image, item.controlSource.image, item.subjectTarget.image,
item.depthControl.ready ? item.depthControl.target.image : VK_NULL_HANDLE,
item.stencilControl.ready ? item.stencilControl.target.image : VK_NULL_HANDLE};
for (const VkImage image : images) {
if (image == VK_NULL_HANDLE) continue;
ImageBarrier(cmd, image, item.aspects, VK_IMAGE_LAYOUT_UNDEFINED,
VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT);
}
}
// (2) The inputs and the sentinels, cleared the wire's way.
for (const ClearPass& clear : clears) {
VkRenderPassBeginInfo begin{VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO};
begin.renderPass = clear.pass;
begin.framebuffer = clear.framebuffer;
begin.renderArea = fullRect;
vkCmdBeginRenderPass(cmd, &begin, VK_SUBPASS_CONTENTS_INLINE);
VkClearAttachment attachment{};
attachment.aspectMask = clear.aspects;
attachment.clearValue.depthStencil = {clear.depth, clear.stencil};
const VkClearRect rect{fullRect, 0, 1};
vkCmdClearAttachments(cmd, 1, &attachment, 1, &rect);
vkCmdEndRenderPass(cmd);
}
// (3) What TransitionWireImage records when the layout already matches - the case of a
// source the application cleared and then resolves.
{
VkMemoryBarrier memory{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
memory.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
memory.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
vkCmdPipelineBarrier(cmd, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1,
&memory, 0, nullptr, 0, nullptr);
}
// (4) THE SUBJECT: the render-pass arm, begun and ended with no draw. Under the test-only
// elideSubject it is not recorded at all: its target keeps the sentinel cleared in (2) and
// the layout it was left in, which is what (6) expects either way.
if (!context.elideSubject) {
for (const FormatWork& item : work) {
VkRenderPassBeginInfo begin{VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO};
begin.renderPass = item.resolvePass;
begin.framebuffer = item.resolveFramebuffer;
begin.renderArea = fullRect;
vkCmdBeginRenderPass(cmd, &begin, VK_SUBPASS_CONTENTS_INLINE);
vkCmdEndRenderPass(cmd);
}
}
// (5) THE CONTROL: the shader arm on the twin source, which TransitionWireImage moves to
// SHADER_READ_ONLY_OPTIMAL first (both aspects of a packed format).
for (const FormatWork& item : work) {
if (!item.depthControl.ready && !item.stencilControl.ready) continue;
ImageBarrier(cmd, item.controlSource.image, item.aspects, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL,
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT);
for (const ControlPass* control : {&item.depthControl, &item.stencilControl}) {
if (!control->ready) continue;
RecordWireShaderResolveDraw(cmd, objects.kit, control->pass, control->pipeline, control->framebuffer,
control->set, {kProbeExtent, kProbeExtent}, fullRect);
}
}
// (6) Every target to TRANSFER_SRC with the barrier ResolveWireDepthStencil puts after its
// resolve (a depth/stencil resolve writes in COLOR_ATTACHMENT_OUTPUT), then one copy per aspect.
const auto copyOut = [&](VkImage image, VkImageAspectFlags aspect, VkDeviceSize offset) {
VkBufferImageCopy copy{};
copy.bufferOffset = offset;
copy.imageSubresource = {aspect, 0, 0, 1};
copy.imageExtent = {kProbeExtent, kProbeExtent, 1};
vkCmdCopyImageToBuffer(cmd, image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, objects.buffer, 1, &copy);
};
const auto toTransfer = [&](VkImage image, VkImageAspectFlags aspects) {
ImageBarrier(cmd, image, aspects, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL,
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_READ_BIT);
};
for (const FormatWork& item : work) {
toTransfer(item.subjectTarget.image, item.aspects);
if (item.aspects & VK_IMAGE_ASPECT_DEPTH_BIT)
copyOut(item.subjectTarget.image, VK_IMAGE_ASPECT_DEPTH_BIT, item.subjectDepthOffset);
if (item.aspects & VK_IMAGE_ASPECT_STENCIL_BIT)
copyOut(item.subjectTarget.image, VK_IMAGE_ASPECT_STENCIL_BIT, item.subjectStencilOffset);
if (item.depthControl.ready) {
toTransfer(item.depthControl.target.image, item.aspects);
copyOut(item.depthControl.target.image, VK_IMAGE_ASPECT_DEPTH_BIT, item.depthControl.offset);
}
if (item.stencilControl.ready) {
toTransfer(item.stencilControl.target.image, item.aspects);
copyOut(item.stencilControl.target.image, VK_IMAGE_ASPECT_STENCIL_BIT, item.stencilControl.offset);
}
}
{
VkBufferMemoryBarrier barrier{VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER};
barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_HOST_READ_BIT;
barrier.srcQueueFamilyIndex = barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = objects.buffer;
barrier.size = VK_WHOLE_SIZE;
vkCmdPipelineBarrier(cmd, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0, 0, nullptr, 1,
&barrier, 0, nullptr);
}
if (vkEndCommandBuffer(cmd) != VK_SUCCESS) return fail("vkEndCommandBuffer failed");
VkFenceCreateInfo fenceInfo{VK_STRUCTURE_TYPE_FENCE_CREATE_INFO};
if (vkCreateFence(device, &fenceInfo, nullptr, &objects.fence) != VK_SUCCESS)
return fail("vkCreateFence failed");
VkSubmitInfo submit{VK_STRUCTURE_TYPE_SUBMIT_INFO};
submit.commandBufferCount = 1;
submit.pCommandBuffers = &cmd;
if (const VkResult result = vkQueueSubmit(context.queue, 1, &submit, objects.fence); result != VK_SUCCESS)
return fail(format("vkQueueSubmit failed (VkResult {})", static_cast<Int>(result)));
const VkResult waited = vkWaitForFences(device, 1, &objects.fence, VK_TRUE, kFenceTimeoutNs);
if (waited == VK_TIMEOUT) {
// The submission may still be executing: destroying anything it names, or idle-waiting
// the device, is the hang the bound exists to prevent. Everything leaks.
leak = true;
measurement.fenceWaitTimedOut = true;
return fail("the probe's fence did not signal within 5 s");
}
if (waited != VK_SUCCESS) return fail(format("vkWaitForFences failed (VkResult {})", static_cast<Int>(waited)));
void* mapped = nullptr;
if (vkMapMemory(device, bufferMemory, 0, VK_WHOLE_SIZE, 0, &mapped) != VK_SUCCESS || mapped == nullptr)
return fail("vkMapMemory failed");
if (!bufferCoherent) {
VkMappedMemoryRange range{VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE};
range.memory = bufferMemory;
range.size = VK_WHOLE_SIZE;
vkInvalidateMappedMemoryRanges(device, 1, &range);
}
const Uint8* bytes = static_cast<const Uint8*>(mapped);
const auto tally = [&](WireDepthResolveAspectReading& aspect, VkDeviceSize offset, Uint32 texelBytes,
Bool depth, VkFormat format, Bool subject) {
Uint32 matches = 0, sentinels = 0, first = 0;
for (Uint32 i = 0; i < kProbeTexels; ++i) {
const Uint8* texel = bytes + offset + static_cast<VkDeviceSize>(i) * texelBytes;
const Uint32 word = depth ? ReadDepthWord(format, texel) : static_cast<Uint32>(*texel);
if (i == 0) first = word;
if (WordDistance(word, aspect.expected) <= aspect.tolerance) ++matches;
if (WordDistance(word, aspect.sentinel) <= aspect.tolerance) ++sentinels;
}
if (subject) {
aspect.renderPassMatches = matches;
aspect.renderPassSentinels = sentinels;
aspect.renderPassFirst = first;
} else {
aspect.shaderRan = true;
aspect.shaderMatches = matches;
aspect.shaderFirst = first;
}
};
for (const FormatWork& item : work) {
WireDepthResolveFormatReading& reading = measurement.formats[item.reading];
if (item.aspects & VK_IMAGE_ASPECT_DEPTH_BIT) {
auto& aspect = reading.depth;
aspect.measured = true;
aspect.texels = kProbeTexels;
aspect.expected = EncodeDepth(item.format, kClearDepth);
aspect.sentinel = EncodeDepth(item.format, kSentinelDepth);
aspect.tolerance = IsFloatDepth(item.format) ? 0u : 1u;
tally(aspect, item.subjectDepthOffset, DepthCopyBytes(item.format), true, item.format, true);
if (item.depthControl.ready)
tally(aspect, item.depthControl.offset, DepthCopyBytes(item.format), true, item.format, false);
}
if (item.aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
auto& aspect = reading.stencil;
aspect.measured = true;
aspect.texels = kProbeTexels;
aspect.expected = kClearStencil;
aspect.sentinel = kSentinelStencil;
aspect.tolerance = 0;
tally(aspect, item.subjectStencilOffset, 1, false, item.format, true);
if (item.stencilControl.ready)
tally(aspect, item.stencilControl.offset, 1, false, item.format, false);
}
}
vkUnmapMemory(device, bufferMemory);
measurement.ran = true;
if (!kitFailure.empty()) measurement.failureReason = kitFailure;
return measurement;
}
WireDepthResolveProbeMeasurement RunWireDepthResolveProbeOnThrowawayDevice(
PFN_vkGetInstanceProcAddr getInstanceProcAddr, VkInstance instance, VkPhysicalDevice physicalDevice,
Uint32 graphicsQueueFamilyIndex, const Vector<VkExtensionProperties>& deviceExtensions,
Bool& renderPassArmAvailable, Bool& shaderStencilExport) {
WireDepthResolveProbeMeasurement measurement;
const auto has = [&](const char* name) {
for (const VkExtensionProperties& extension : deviceExtensions)
if (std::strcmp(extension.extensionName, name) == 0) return true;
return false;
};
renderPassArmAvailable = has(VK_KHR_CREATE_RENDERPASS_2_EXTENSION_NAME) &&
has(VK_KHR_DEPTH_STENCIL_RESOLVE_EXTENSION_NAME);
shaderStencilExport = has(VK_EXT_SHADER_STENCIL_EXPORT_EXTENSION_NAME);
if (!renderPassArmAvailable) {
measurement.failureReason = "no VK_KHR_create_renderpass2 + VK_KHR_depth_stencil_resolve";
return measurement;
}
VkPhysicalDeviceProperties properties{};
vkGetPhysicalDeviceProperties(physicalDevice, &properties);
auto getProperties2 = reinterpret_cast<PFN_vkGetPhysicalDeviceProperties2>(
getInstanceProcAddr(instance, "vkGetPhysicalDeviceProperties2"));
if (getProperties2 == nullptr)
getProperties2 = reinterpret_cast<PFN_vkGetPhysicalDeviceProperties2>(
getInstanceProcAddr(instance, "vkGetPhysicalDeviceProperties2KHR"));
const auto createDevice =
reinterpret_cast<PFN_vkCreateDevice>(getInstanceProcAddr(instance, "vkCreateDevice"));
if (properties.apiVersion < VK_API_VERSION_1_1 || getProperties2 == nullptr || createDevice == nullptr) {
renderPassArmAvailable = false;
measurement.failureReason = "a pre-1.1 device, or no vkGetPhysicalDeviceProperties2 / vkCreateDevice";
return measurement;
}
VkPhysicalDeviceDepthStencilResolveProperties resolveProperties{
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES};
VkPhysicalDeviceProperties2 properties2{VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2};
properties2.pNext = &resolveProperties;
getProperties2(physicalDevice, &properties2);
// The renderer's own extension set for the two arms (VulkanRenderer.cpp,
// CreateLogicalDeviceAndQueues), nothing more.
const Float priority = 1.0f;
VkDeviceQueueCreateInfo queueInfo{VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO};
queueInfo.queueFamilyIndex = graphicsQueueFamilyIndex;
queueInfo.queueCount = 1;
queueInfo.pQueuePriorities = &priority;
Vector<const char*> extensions{VK_KHR_CREATE_RENDERPASS_2_EXTENSION_NAME,
VK_KHR_DEPTH_STENCIL_RESOLVE_EXTENSION_NAME};
if (shaderStencilExport) extensions.push_back(VK_EXT_SHADER_STENCIL_EXPORT_EXTENSION_NAME);
VkDeviceCreateInfo deviceInfo{VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO};
deviceInfo.queueCreateInfoCount = 1;
deviceInfo.pQueueCreateInfos = &queueInfo;
deviceInfo.enabledExtensionCount = static_cast<Uint32>(extensions.size());
deviceInfo.ppEnabledExtensionNames = extensions.data();
VkDevice device = VK_NULL_HANDLE;
if (const VkResult result = createDevice(physicalDevice, &deviceInfo, nullptr, &device);
result != VK_SUCCESS || device == VK_NULL_HANDLE) {
measurement.failureReason = format("vkCreateDevice failed (VkResult {})", static_cast<Int>(result));
return measurement;
}
WireDepthResolveProbeContext context;
context.physicalDevice = physicalDevice;
context.device = device;
vkGetDeviceQueue(device, graphicsQueueFamilyIndex, 0, &context.queue);
context.queueFamilyIndex = graphicsQueueFamilyIndex;
context.createRenderPass2 =
reinterpret_cast<PFN_vkCreateRenderPass2>(vkGetDeviceProcAddr(device, "vkCreateRenderPass2KHR"));
context.depthResolveModes = resolveProperties.supportedDepthResolveModes;
context.stencilResolveModes = resolveProperties.supportedStencilResolveModes;
context.shaderStencilExport = shaderStencilExport;
measurement = RunWireDepthResolveProbe(context);
// A timed-out probe leaked its children on purpose; destroying their device would block
// exactly where the bound says not to.
if (!measurement.fenceWaitTimedOut) vkDestroyDevice(device, nullptr);
return measurement;
}
} // namespace MobileGL::MG_Backend::DirectVulkan
#endif
@@ -0,0 +1,121 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WireDepthResolveProbe.h
// Copyright (c) 2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include "WireDepthResolveArm.h"
#include <Config.h>
#include <Includes.h>
// P7 gate 5 (g5-msprobe): the Vulkan half of the multisample depth/stencil resolve probe, and the
// two resolve arms it measures.
//
// THE ARMS LIVE HERE, NOT ONLY IN THE RENDERER. A probe that rebuilt "something like" the wire's
// render pass would measure its own copy; the render-pass description, the shader pass's modules,
// pipeline and draw, and the view shapes both arms attach are defined once below and called by
// ResolveWireDepthStencil (WireFramebuffer.inc), ResolveWireDepthStencilWithShader
// (WireMultisampleResolve.inc) and the probe alike.
//
// Two callers: the Magma renderer at device bring-up (server side, once per process, only where
// the render-pass arm exists at all - VulkanRenderer::ArmWireDepthResolveOrder) and the POST's
// "Known Driver Bugs" section (MG_Util/SelfTest/DriverBugProbes.cpp) on a throwaway device.
//
// Disaggregated builds only: the monolith never records either arm, and the pull build's image
// must not change (P7 gate G1).
#if MOBILEGL_BUILD_DISAGGREGATED
namespace MobileGL::MG_Backend::DirectVulkan {
// Every aspect a wire depth/stencil format has.
VkImageAspectFlags WireDepthStencilFormatAspects(VkFormat format);
// A 2D view of one level/layer of a depth/stencil image naming `aspects`. The render-pass arm
// attaches both images with every aspect of the format; the shader arm SAMPLES its source
// through a view of exactly one aspect (a depth/stencil image can only be sampled that way) and
// attaches its target with every aspect.
VkImageViewCreateInfo WireDepthStencilViewInfo(VkImage image, VkFormat format, VkImageAspectFlags aspects,
Uint32 level, Uint32 layer);
// THE RENDER-PASS ARM: attachment 0 the multisample source (LOAD/STORE), attachment 1 the
// single-sample resolve target (DONT_CARE/STORE), both in DEPTH_STENCIL_ATTACHMENT_OPTIMAL, one
// subpass whose VkSubpassDescriptionDepthStencilResolve resolves every aspect of the format
// with VK_RESOLVE_MODE_SAMPLE_ZERO_BIT. The caller begins and ends it with NO draw.
VkResult CreateWireDepthStencilResolvePass(PFN_vkCreateRenderPass2 createRenderPass2, VkDevice device,
VkFormat format, VkSampleCountFlagBits samples, VkRenderPass* pass);
// THE SHADER ARM's device objects that do not depend on the format.
struct WireShaderResolveKit {
VkShaderModule vertex = VK_NULL_HANDLE;
VkShaderModule depthFragment = VK_NULL_HANDLE;
// Only where VK_EXT_shader_stencil_export is enabled: its SPIR-V declares StencilExportEXT.
VkShaderModule stencilFragment = VK_NULL_HANDLE;
VkDescriptorSetLayout descriptorLayout = VK_NULL_HANDLE;
VkPipelineLayout pipelineLayout = VK_NULL_HANDLE;
VkSampler nearest = VK_NULL_HANDLE;
};
// Creates every member; the first failure is returned and the members made so far stay set,
// for DestroyWireShaderResolveKit.
VkResult CreateWireShaderResolveKit(VkDevice device, Bool shaderStencilExport, WireShaderResolveKit& kit);
void DestroyWireShaderResolveKit(VkDevice device, WireShaderResolveKit& kit);
// One (format, aspect)'s single-attachment render pass: DONT_CARE/STORE on both aspects,
// DEPTH_STENCIL_ATTACHMENT_OPTIMAL throughout.
VkResult CreateWireShaderResolveRenderPass(VkDevice device, VkFormat format, VkRenderPass* pass);
// Its pipeline: full-screen triangle, gl_FragDepth (ALWAYS + write) or gl_FragStencilRefARB
// (REPLACE everywhere), viewport and scissor dynamic. VK_ERROR_FEATURE_NOT_PRESENT when the kit
// has no fragment module for `aspect`.
VkResult CreateWireShaderResolvePipeline(VkDevice device, const WireShaderResolveKit& kit, VkRenderPass pass,
VkImageAspectFlags aspect, VkPipeline* pipeline);
// Records the draw: the render area and the scissor are `rect`, the viewport is the WHOLE
// attachment (gl_FragCoord is what the fragment stage fetches by), one triangle.
void RecordWireShaderResolveDraw(VkCommandBuffer commandBuffer, const WireShaderResolveKit& kit,
VkRenderPass pass, VkPipeline pipeline, VkFramebuffer framebuffer,
VkDescriptorSet descriptorSet, VkExtent2D attachmentExtent, VkRect2D rect);
// What the probe runs on. The device must have been created with VK_KHR_create_renderpass2 and
// VK_KHR_depth_stencil_resolve (or Vulkan 1.2 core), and with VK_EXT_shader_stencil_export where
// `shaderStencilExport` says so. The queue must be otherwise idle for the probe's duration.
struct WireDepthResolveProbeContext {
VkPhysicalDevice physicalDevice = VK_NULL_HANDLE;
VkDevice device = VK_NULL_HANDLE;
VkQueue queue = VK_NULL_HANDLE;
Uint32 queueFamilyIndex = 0;
PFN_vkCreateRenderPass2 createRenderPass2 = nullptr;
VkResolveModeFlags depthResolveModes = 0;
VkResolveModeFlags stencilResolveModes = 0;
Bool shaderStencilExport = false;
// TEST ONLY (MGITEST_MAGMA_DEPTH_RESOLVE_PROBE=elide-subject, WireDepthResolveArm.h): record
// NO render-pass resolve, so every subject target keeps its sentinel - the Adreno 830's
// reading - while the shader control runs as always. Everything else is the real probe.
Bool elideSubject = false;
};
// Per format the device can host as a 4x multisample depth/stencil attachment AND for which
// ResolveWireDepthStencil would take the render-pass arm (D16, X8_D24, D32F, D24S8, D32FS8, S8):
// - two identical 4x4 4x multisample sources (production usage), each cleared to depth 0.25 /
// stencil 0x5A the way the wire clears (a LOAD pass with vkCmdClearAttachments);
// - single-sample resolve targets (the production scratch usage) pre-filled with the sentinel
// depth 0.75 / stencil 0xA5;
// - SUBJECT: the render-pass arm's exact pass, begun and ended with no draw;
// - CONTROL: the shader arm's exact pass on the other source, one target per aspect (stencil
// only with VK_EXT_shader_stencil_export);
// - every target copied aspect by aspect into a host-visible buffer and compared texel by texel.
// One submission, a bounded fence wait (on timeout every object is leaked and the measurement
// says so - the caller must then not destroy or idle-wait the device), teardown on every other
// path. Never touches renderer state.
WireDepthResolveProbeMeasurement RunWireDepthResolveProbe(const WireDepthResolveProbeContext& context);
// The POST's entry: creates a throwaway device on `physicalDevice` with exactly the extensions
// the renderer enables for the two arms, runs the probe on its graphics queue and destroys the
// device (leaks it on a fence timeout). `renderPassArmAvailable` is false - and nothing is run -
// when the device lacks VK_KHR_create_renderpass2 / VK_KHR_depth_stencil_resolve; the shader
// pass is then the only arm and the probe has nothing to choose between.
// `shaderStencilExport` reports whether VK_EXT_shader_stencil_export was available.
WireDepthResolveProbeMeasurement RunWireDepthResolveProbeOnThrowawayDevice(
PFN_vkGetInstanceProcAddr getInstanceProcAddr, VkInstance instance, VkPhysicalDevice physicalDevice,
Uint32 graphicsQueueFamilyIndex, const Vector<VkExtensionProperties>& deviceExtensions,
Bool& renderPassArmAvailable, Bool& shaderStencilExport);
} // namespace MobileGL::MG_Backend::DirectVulkan
#endif
@@ -11,9 +11,13 @@
return MagmaProgramSource(handle, records[index]);
}
// Through the same seam as MagmaWireFatal, and for the same reason (P7 wave 0). The family
// stays RoleViolation: reaching the legacy buffer arm inside a Magma server is a role
// invariant the two halves must hold, not a verb the wire failed to migrate.
[[noreturn]] static void WireBufferLegacyFatal() {
MGLOG_F("MGPipe: Fatal{RoleViolation, \"buffer-legacy-arm\"} (Magma P7 buffer consumer)");
std::abort();
MG_Pipe::MGPipeSessionFail(
MG_Pipe::MGPipeFatalFamily::RoleViolation,
"MGPipe: Fatal{RoleViolation, \"buffer-legacy-arm\"} (Magma P7 buffer consumer)");
}
static const MG_Pipe::MGPStreamOutputBegin& WireXfbSpan() {
@@ -199,8 +203,14 @@
Flags<DrawSetupAspect> aspects, const DrawCmdParam& drawParams, const IndexBufferView* indices) {
auto& state = MG_Pipe::MGPipeApplier();
const auto program = WireProgramSource(state.BoundShaderCso);
if (!program.GetLinkStatus() || !program.GetSpirvStatus()) return false;
if (aspects & DrawSetupAspect::IndirectDrawBuffer)
if (!program.GetLinkStatus() || !program.GetSpirvStatus()) {
MGL_WIRE_DECLINE_AT(ProgramLinkOrSpirv, "program is not linked or carries no SPIR-V");
return false;
}
// P8-D: the wire arm's own indirect draw (DrawWireIndirectNative) is the one caller that
// may pass this aspect; a monolith indirect entry point reached here is still the role
// violation.
if ((aspects & DrawSetupAspect::IndirectDrawBuffer) && !m_wireNativeIndirectDraw)
WireBufferLegacyFatal();
MakeXfbWritesVisible();
const auto* fbo = state.DrawFramebuffer();
@@ -220,10 +230,16 @@
flags |= ProgramFactory::CompileOptionBit::ZeroBaseVertex;
programObject = &m_programFactory->GetOrCreateProgram(program, flags);
}
if (programObject->pointSizeCapabilityUnsupported) return false;
if (programObject->pointSizeCapabilityUnsupported) {
MGL_WIRE_DECLINE_AT(PointSizeCapability, "the program writes gl_PointSize and the device cannot express it");
return false;
}
const auto& programObj = *programObject;
m_currentDrawXfbCaptureDeclined = programObj.xfbCaptureDeclined;
if (!m_uniformManager->PrepareWireTextureResources(program, programObj)) return false;
if (!m_uniformManager->PrepareWireTextureResources(program, programObj)) {
MGL_WIRE_DECLINE_AT(TextureResourcesUnprepared, "a sampled/storage image could not be made ready; see the Tex* tally");
return false;
}
// Lazy image uploads/promotions can submit the old recording. Resolve
// ALL attachments before descriptor binding captures the current cmd.
for (Uint32 i = 0; i < MG_Pipe::kMGPipeMaxColorAttachments; ++i) {
@@ -241,37 +257,106 @@
const auto& vao = state.VertexElementsCsos[vaoHandle.Slot];
if (!vao.Live || vao.Gen != vaoHandle.Gen) MagmaWireFatal("vertex-elements-generation");
VertexInputStateFactory::BackendVertexInputState input{};
// P7 wave 2 package C, CONTRACT-P7 §3.2: THE `vertex-layout` ROW IS RETIRED AND SPLIT.
//
// A false is now ONE reason and only one - `buffer-window`, an enabled attribute the
// program reads whose slot lies outside the window set_vertex_buffers published. That
// is a disagreement about what crossed, not about what this device can express, so it
// keeps a named Fatal through the hook (MGPipeSessionFail -> Session::Fail, which is
// what rule I asks of a refusal that must stay a refusal). It is the check that caught
// the Redmi VAO hash collision fixed in 376c04be, which is why it is not softened.
// The P7 marker is gone from the detail string with the abort: this is a permanent
// protocol error, not a shape the split path has yet to migrate, so it must not keep
// counting against the census of unmigrated refusals (scripts/ci/fatal_census.py
// counts the marker literally, which is also why no comment in this package spells it).
if (!m_vertexInputStateFactory->BuildWireVertexInput(vao, state,
programObj.activeVertexInputLocationMask, input)) MagmaWireFatal("vertex-layout@P7");
programObj.activeVertexInputLocationMask, input))
MagmaWireFatal("vertex-layout-buffer-window");
// The seven device/format reasons are now MASKED OUT of the vertex input state, and
// this is the monolith arm's pre-flight for them, statement for statement
// (VulkanRenderer.cpp's SetupDraw: "Fail loudly rather than render wrong pixels"). An
// enabled array with no usable VkFormat never reaches the vertex input state, which
// makes it indistinguishable from a DISABLED one - and a disabled one is answered
// below by uploading the current attribute value, so without this check the draw would
// quietly feed the shader a constant in place of the application's vertex data.
//
// A DECLINE, NOT A FATAL (rule I (a)): the draw is skipped, the reason is named once,
// and the observable is the monolith lane's for the same VAO on the same device.
// BuildWireVertexInput only considers attributes the program reads, so a masked bit is
// always an active one and this predicate is the whole of the difference.
if ((input.unsupportedAttribMask & programObj.activeVertexInputLocationMask) != 0) {
MGLOG_E_ONCE("Magma wire draw skipped: program=%u reads vertex attribute location "
"mask 0x%x whose enabled array has no supported vertex format",
program.GetExternalIndex(),
input.unsupportedAttribMask & programObj.activeVertexInputLocationMask);
WireDeclineTally::Count(WireDeclineSite::VertexFormatUnsupported);
return false;
}
Vector<BufferSlice> vertexSlices(input.bindings.size());
for (SizeT binding = 0; binding < input.bindings.size(); ++binding) {
const Uint32 location = input.bindingAttributeLocations[binding];
const auto& attr = vao.Attributes[location];
const auto& resource = state.VertexBuffers[location];
if (MG_Pipe::MGPipeHandleIsNull(resource.Res)) MagmaWireFatal("client-vertex-array@P8");
// P8-F: A RECORD INVARIANT, NOT AN UNMIGRATED VERB (it carried a P8 phase tag).
// The client turns every enabled client-memory array into an owned buffer before the
// draw record is built, or refuses the draw itself (MG_Impl/Pipe/OwnedDrawInputs.h),
// so an attribute the program reads with no buffer handle is a foreign peer's record.
if (MG_Pipe::MGPipeHandleIsNull(resource.Res))
MG_Pipe::MGPipeSessionFail(MG_Pipe::MGPipeFatalFamily::ProtocolCorruption,
"MGPipe: Fatal{ProtocolCorruption, \"Magma:client-vertex-array\"} location=%u: an "
"attribute the program reads carries no buffer handle", location);
BufferSlice slice{};
if (!m_bufferManager.AcquireWireSlice(BufferKind::Vertex, resource.Res, slice))
MagmaWireFatal("vertex-buffer-resource");
const VkDeviceSize offset = input.bindingBaseOffsets[binding];
const auto type = static_cast<DataType>(attr.Type);
const SizeT elementSize = VertexInputStateFactory::GetAttributeByteSize(type, attr.Size, attr.IsBgra);
if (offset > slice.size || elementSize > slice.size - offset) return false;
if (offset > slice.size || elementSize > slice.size - offset) {
MGL_WIRE_DECLINE_AT(VertexAttributeWindow, "an enabled attribute's first element lies outside its buffer");
return false;
}
const auto conversion = input.bindingConversions[binding];
if (conversion != VertexInputStateFactory::VertexStreamConversion::None) {
const SizeT stride = static_cast<SizeT>(attr.Stride);
SizeT count = stride ? 1 + (static_cast<SizeT>(slice.size - offset) - elementSize) / stride : 1;
// Direct non-indexed draws bound the conversion; indexed/multi-draw may
// name arbitrary vertices, so preserve the entire valid source stream.
if (stride && !(aspects & DrawSetupAspect::IndexBuffer) && !resource.Divisor)
// name arbitrary vertices, so preserve the entire valid source stream. So may an
// indirect draw (P8-D): its vertex range is in GPU words nobody read.
if (stride && !(aspects & DrawSetupAspect::IndexBuffer) &&
!(aspects & DrawSetupAspect::IndirectDrawBuffer) && !resource.Divisor)
count = std::min<SizeT>(count, static_cast<SizeT>(drawParams.firstVertex) + drawParams.vertexCount);
if (!count) return false;
if (!count) {
MGL_WIRE_DECLINE_AT(VertexStreamCountZero, "the converted vertex stream would hold no elements");
return false;
}
Vector<Uint8> source((count - 1) * stride + elementSize);
if (!m_bufferManager.ReadWireBuffer(resource.Res, offset, source.size(), source.data())) return false;
if (!m_bufferManager.ReadWireBuffer(resource.Res, offset, source.size(), source.data())) {
MGL_WIRE_DECLINE_AT(VertexStreamRead, "the vertex bytes to convert could not be read back");
return false;
}
Vector<Uint8> converted;
if (!ConvertWireVertexStream(attr, conversion, source.data(), stride, count, converted))
MagmaWireFatal("vertex-format-conversion@P7");
// P7 wave 2 package C, CONTRACT-P7 §3.2 `vertex-format-conversion`: DECLINE.
// The converter is asked only for a conversion BuildWireVertexInput already
// chose, so a refusal here means the two disagree about what is convertible -
// a MobileGL-internal defect, not a peer's malformed record and not a verb the
// split path has yet to migrate. Killing the session for it told the peer
// "protocol corruption" about a bug on this side; skipping the draw and naming
// it once is the honest answer and is what the row asks for. The packed
// converter that would make the refusal unreachable is the row's optional half
// and is not taken here.
if (!ConvertWireVertexStream(attr, conversion, source.data(), stride, count, converted)) {
MGLOG_E_ONCE("Magma wire draw skipped: no vertex stream converter for "
"location=%u type=%u size=%u conversion=%d",
location, attr.Type, static_cast<Uint32>(attr.Size),
static_cast<Int>(conversion));
WireDeclineTally::Count(WireDeclineSite::VertexStreamConverter);
return false;
}
if (!m_bufferManager.UploadTransient(BufferKind::Vertex, m_frameContext.GetCurrentFrameIndex(),
converted.data(), converted.size(), 16, vertexSlices[binding])) return false;
converted.data(), converted.size(), 16, vertexSlices[binding])) {
MGL_WIRE_DECLINE_AT(VertexTransientUpload, "the converted vertex stream did not fit the transient arena");
return false;
}
} else {
slice.offset += offset;
slice.size -= offset;
@@ -282,9 +367,15 @@
BufferSlice indexSlice{};
VkIndexType indexType = VK_INDEX_TYPE_MAX_ENUM;
if (aspects & DrawSetupAspect::IndexBuffer) {
if (!indices) return false;
if (!indices) {
MGL_WIRE_DECLINE_AT(IndexViewMissing, "an indexed draw arrived with no index view");
return false;
}
const SizeT width = MG_Util::GetGLTypeSize(indices->indexType);
if (width != 1 && width != 2 && width != 4) return false;
if (width != 1 && width != 2 && width != 4) {
MGL_WIRE_DECLINE_AT(IndexTypeWidth, "the index type is not 1, 2 or 4 bytes wide");
return false;
}
indexType = width == 4 ? VK_INDEX_TYPE_UINT32 : width == 2 ? VK_INDEX_TYPE_UINT16 : VK_INDEX_TYPE_UINT8;
const auto& restart = MGB_CTX->GetRenderStateParameters();
const Uint32 fixed = MG_Util::FixedRestartIndexForGLType(indices->indexType);
@@ -300,10 +391,16 @@
MagmaWireFatal("index-buffer-resource");
const Uint64 offset = state.IndexBuffer.Offset + indices->indexByteOffset;
if (offset < state.IndexBuffer.Offset || offset > indexSlice.size ||
indices->indexByteSize > indexSlice.size - offset) return false;
indices->indexByteSize > indexSlice.size - offset) {
MGL_WIRE_DECLINE_AT(IndexWindow, "the index range lies outside its buffer");
return false;
}
if (replaceRestart || expandByte || offset % width) {
copied.resize(indices->indexByteSize);
if (!m_bufferManager.ReadWireBuffer(state.IndexBuffer.Res, offset, copied.size(), copied.data())) return false;
if (!m_bufferManager.ReadWireBuffer(state.IndexBuffer.Res, offset, copied.size(), copied.data())) {
MGL_WIRE_DECLINE_AT(IndexStreamRead, "the indices to rewrite could not be read back");
return false;
}
bytes = copied.data();
}
indexSlice.offset += offset;
@@ -327,9 +424,15 @@
if (expandByte) indexType = VK_INDEX_TYPE_UINT16;
}
if (!m_bufferManager.UploadTransient(BufferKind::Index, m_frameContext.GetCurrentFrameIndex(),
bytes, uploadSize, 16, indexSlice)) return false;
bytes, uploadSize, 16, indexSlice)) {
MGL_WIRE_DECLINE_AT(IndexTransientUpload, "the rewritten indices did not fit the transient arena");
return false;
}
}
if (!indexSlice.IsValid()) {
MGL_WIRE_DECLINE_AT(IndexSliceInvalid, "no usable index slice was resolved");
return false;
}
if (!indexSlice.IsValid()) return false;
}
if (!frame.isCommandRecording) m_frameContext.BeginCommandRecording();
@@ -337,7 +440,10 @@
// Resolve it before constructing framebuffer views so both roles of that image
// name the same allocation even on its first storage-image use.
if (!m_uniformManager->BindProgramUniformBuffers(frame.commandBuffer, program, programObj,
m_frameContext.GetCurrentFrameIndex(), VK_PIPELINE_BIND_POINT_GRAPHICS)) return false;
m_frameContext.GetCurrentFrameIndex(), VK_PIPELINE_BIND_POINT_GRAPHICS)) {
MGL_WIRE_DECLINE_AT(UniformBufferBinding, "the program's uniform/storage bindings could not be bound");
return false;
}
Vector<VkAttachmentDescription> attachments;
Vector<VkImageViewCreateInfo> viewInfos;
@@ -409,7 +515,10 @@
depth.attachment = add(fbo->Depth, VK_IMAGE_ASPECT_DEPTH_BIT);
else if (fbo->Stencil.Kind != MG_Pipe::kMGPipeSurfaceKindNone)
depth.attachment = add(fbo->Stencil, VK_IMAGE_ASPECT_STENCIL_BIT);
if (attachments.empty()) return false;
if (attachments.empty()) {
MGL_WIRE_DECLINE_AT(NoAttachments, "the draw framebuffer resolved to no attachments");
return false;
}
passKey.width = extent.width;
passKey.height = extent.height;
passKey.layers = layers;
@@ -466,7 +575,10 @@
const Uint64 compatibilityId = m_wireRenderPassCompatibility.Intern(pipelinePassKey);
const auto pipeline = GetOrCreatePipelineWithInput(mode, program, programObj, flags, input, pass,
ResolvePrimitiveRestartEnable(aspects, indices), compatibilityId);
if (pipeline == VK_NULL_HANDLE) return false;
if (pipeline == VK_NULL_HANDLE) {
MGL_WIRE_DECLINE_AT(PipelineUnavailable, "no graphics pipeline could be created for this draw");
return false;
}
for (Uint32 i = 0; i < vertexSlices.size(); ++i)
vkCmdBindVertexBuffers(frame.commandBuffer, i, 1, &vertexSlices[i].buffer, &vertexSlices[i].offset);
Uint32 binding = static_cast<Uint32>(vertexSlices.size());
@@ -477,14 +589,23 @@
const void* data = nullptr;
VkDeviceSize size = 0;
if (!TryGetCurrentVertexAttributeUploadPayload(MGB_CTX->GetCurrentVertexAttribute(location),
programObj.vertexInputTypes[location], format, data, size)) return false;
programObj.vertexInputTypes[location], format, data, size)) {
MGL_WIRE_DECLINE_AT(CurrentAttributePayload, "a disabled array's current attribute value has no upload payload");
return false;
}
BufferSlice slice{};
if (!m_bufferManager.UploadTransient(BufferKind::Vertex, m_frameContext.GetCurrentFrameIndex(),
data, size, 16, slice)) return false;
data, size, 16, slice)) {
MGL_WIRE_DECLINE_AT(CurrentAttributeUpload, "the current attribute value did not fit the transient arena");
return false;
}
vkCmdBindVertexBuffers(frame.commandBuffer, binding++, 1, &slice.buffer, &slice.offset);
}
if (indexSlice.IsValid()) vkCmdBindIndexBuffer(frame.commandBuffer, indexSlice.buffer, indexSlice.offset, indexType);
if (!VkRenderPassManager::BeginRenderPass(frame.commandBuffer, pass)) return false;
if (!VkRenderPassManager::BeginRenderPass(frame.commandBuffer, pass)) {
MGL_WIRE_DECLINE_AT(BeginRenderPass, "the render pass for this draw could not be begun");
return false;
}
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
ApplyDynamicDrawStateTail(frame, pass.extent, fbo->IsDefault,
ResolveDrawViewportCount(programObj.writesViewportIndexBuiltin));
@@ -492,19 +613,237 @@
return true;
}
// ---- P8-D: THE WIRE ARM'S INDIRECT DRAWS READ THEIR WORDS ON THE GPU ----------------------
//
// The shape the monolith arm records (VulkanRenderer.cpp's MultiDraw{Elements,Arrays}Indirect
// and MultiDrawElementsIndirectCount), with the wire store's own VkBuffer in place of the
// frontend buffer's resident slice. Before this, every wire indirect draw read its commands
// through ReadWireBuffer and expanded them on the CPU, and a store a shader had written made
// that read a whole-GPU wait (WaitForWireBufferHostAccess) per draw.
//
// THE ONE FALSE: a COUNT form this device cannot issue - the monolith arm's own condition for
// vkCmdDrawIndexedIndirectCount (MultiDrawElementsIndirectCount), extended to the non-indexed
// twin the monolith never loads. Everything else is issued here or declined by name.
Bool VulkanRenderer::DrawWireIndirectNative(GLenum mode, GLenum type, Uint64 offset, GLsizei drawcount,
GLsizei stride, Bool indexed, Bool counted, Uint64 countOffset) {
if (drawcount <= 0) return true;
if (counted) {
const auto countEntry = indexed ? s_vkCmdDrawIndexedIndirectCount : s_vkCmdWireDrawIndirectCount;
if (!m_drawIndirectCountExtensionEnabled || countEntry == nullptr ||
(!m_multiDrawIndirectFeatureEnabled && drawcount != 1))
return false;
}
const auto& state = MG_Pipe::MGPipeApplier();
// GL's command sizes (GL 4.6 10.3.10), which are also Vulkan's.
const Uint64 commandSize = indexed ? 5 * sizeof(Uint32) : 4 * sizeof(Uint32);
if (stride == 0) stride = static_cast<GLsizei>(commandSize);
BufferSlice commands{};
if (MG_Pipe::MGPipeHandleIsNull(state.VerbIndirectBuffer) ||
!m_bufferManager.AcquireWireSlice(BufferKind::Indirect, state.VerbIndirectBuffer, commands)) {
MGL_WIRE_DECLINE_AT(IndirectBufferUnbound, "an indirect draw names no live indirect buffer store");
return true;
}
const Uint64 span = static_cast<Uint64>(stride) * static_cast<Uint64>(drawcount - 1) + commandSize;
// GL refuses a stride that is not a multiple of four (INVALID_VALUE), and Vulkan requires
// it for drawCount > 1; a peer's record that breaks either is declined, not issued.
if (stride < static_cast<GLsizei>(commandSize) || (stride & 3) != 0 || (offset & 3u) != 0 ||
offset > commands.size || span > commands.size - offset) {
MGL_WIRE_DECLINE_AT(IndirectCommandRange, "the indirect commands lie outside their store or are misaligned");
return true;
}
BufferSlice parameters{};
if (counted) {
if (MG_Pipe::MGPipeHandleIsNull(state.VerbIndirectParameterBuffer) ||
!m_bufferManager.AcquireWireSlice(BufferKind::Indirect, state.VerbIndirectParameterBuffer, parameters) ||
(countOffset & 3u) != 0 || countOffset > parameters.size || parameters.size - countOffset < sizeof(Uint32)) {
MGL_WIRE_DECLINE_AT(IndirectCountRange, "the indirect draw count lies outside its parameter store");
return true;
}
}
// The index view is the WHOLE bound element store past its binding offset, as the monolith
// arm binds it: the commands name their own firstIndex, which only the GPU reads.
IndexBufferView indices{};
DrawCmdParam vertexRange{};
vertexRange.instanceCount = 1;
Flags<DrawSetupAspect> aspects = DrawSetupAspect::IndirectDrawBuffer;
if (indexed) {
const SizeT width = MG_Util::GetGLTypeSize(type);
BufferSlice elements{};
if (width == 0 || MG_Pipe::MGPipeHandleIsNull(state.IndexBuffer.Res) ||
!m_bufferManager.AcquireWireSlice(BufferKind::Index, state.IndexBuffer.Res, elements) ||
state.IndexBuffer.Offset > elements.size) {
MGL_WIRE_DECLINE_AT(IndirectIndexStore, "an indexed indirect draw names no usable element store");
return true;
}
indices.indexType = type;
indices.indexByteOffset = 0;
indices.indexByteSize = static_cast<SizeT>(elements.size - state.IndexBuffer.Offset);
vertexRange.vertexCount = static_cast<Uint32>(indices.indexByteSize / width);
aspects |= DrawSetupAspect::IndexBuffer;
}
// A shader write the store was marked with (MarkWireBufferGpuWritten) is made visible to
// the indirect read here, once per write, so a program that omits GL_COMMAND_BARRIER_BIT
// is not left racing a read the old CPU path used to take after a whole-GPU wait. The
// application's own glMemoryBarrier is BuildMemoryBarrierForGlBarriers, as on the monolith
// arm; transfer writes carry their own MEMORY_READ after-barrier (VkBufferManager.cpp).
const Bool commandsWritten = m_bufferManager.TakeWireIndirectReadBarrier(state.VerbIndirectBuffer);
const Bool countWritten = counted && m_bufferManager.TakeWireIndirectReadBarrier(state.VerbIndirectParameterBuffer);
if (commandsWritten || countWritten) {
RetireWireDrawPass();
auto& barrierFrame = m_frameContext.GetCurrent();
if (!barrierFrame.isCommandRecording) m_frameContext.BeginCommandRecording();
if (VkRenderPassManager::GetActiveRenderPass() != nullptr)
VkRenderPassManager::EndRenderPass(barrierFrame.commandBuffer);
VkMemoryBarrier barrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
barrier.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT;
vkCmdPipelineBarrier(barrierFrame.commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT, 0, 1, &barrier, 0, nullptr, 0, nullptr);
if (MG_Util::PipeStats::Enabled())
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireIndirectBarriers, 1);
}
auto& frame = m_frameContext.GetCurrent();
m_wireNativeIndirectDraw = true;
const Bool ready = SetupDraw(frame, mode, aspects, vertexRange, indexed ? &indices : nullptr);
m_wireNativeIndirectDraw = false;
if (!ready) return true;
MOBILEGL_ASSERT(frame.isCommandRecording, "%s: frame recording was not started", __func__);
const VkCommandBuffer commandBuffer = frame.commandBuffer;
const VkDeviceSize first = commands.offset + offset;
const Uint32 count = static_cast<Uint32>(drawcount);
const Uint32 byteStride = static_cast<Uint32>(stride);
if (counted) {
const auto countEntry = indexed ? s_vkCmdDrawIndexedIndirectCount : s_vkCmdWireDrawIndirectCount;
countEntry(commandBuffer, commands.buffer, first, parameters.buffer, parameters.offset + countOffset,
count, byteStride);
} else if (count == 1 || (!m_multiDrawForceUnrollIndirect && m_multiDrawIndirectFeatureEnabled)) {
// maxDrawIndirectCount bounds one call; GL's drawcount does not.
const Uint32 limit = std::max<Uint32>(1u, m_physicalDevice.properties.limits.maxDrawIndirectCount);
for (Uint32 base = 0; base < count; base += limit) {
const Uint32 chunk = std::min(count - base, limit);
const VkDeviceSize at = first + static_cast<VkDeviceSize>(base) * byteStride;
if (indexed) vkCmdDrawIndexedIndirect(commandBuffer, commands.buffer, at, chunk, byteStride);
else vkCmdDrawIndirect(commandBuffer, commands.buffer, at, chunk, byteStride);
}
} else {
// No multiDrawIndirect: one indirect draw per command, still read by the GPU
// (the monolith arm's own fallback).
for (Uint32 i = 0; i < count; ++i) {
const VkDeviceSize at = first + static_cast<VkDeviceSize>(i) * byteStride;
if (indexed) vkCmdDrawIndexedIndirect(commandBuffer, commands.buffer, at, 1, 0);
else vkCmdDrawIndirect(commandBuffer, commands.buffer, at, 1, 0);
}
}
if (MG_Util::PipeStats::Enabled())
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireIndirectNativeDraws, 1);
return true;
}
void VulkanRenderer::DispatchWireCompute(GLuint x, GLuint y, GLuint z) {
RetireWireDrawPass();
auto& frame = m_frameContext.GetCurrent();
const auto program = WireProgramSource(MG_Pipe::MGPipeApplier().DispatchProgram);
if (!program.GetLinkStatus() || !program.GetSpirvStatus()) return;
if (!program.GetLinkStatus() || !program.GetSpirvStatus()) {
MGL_WIRE_DECLINE_AT(ComputeProgramLinkOrSpirv, "the compute program is not linked or carries no SPIR-V");
return;
}
const auto& programObj = m_programFactory->GetOrCreateProgram(program, {});
if (!m_uniformManager->PrepareWireTextureResources(program, programObj)) return;
if (!m_uniformManager->PrepareWireTextureResources(program, programObj)) {
MGL_WIRE_DECLINE_AT(ComputeTextureResourcesUnprepared, "a sampled/storage image could not be made ready for the dispatch");
return;
}
const auto pipeline = GetOrCreateComputePipeline(programObj);
if (!pipeline) return;
if (!pipeline) {
MGL_WIRE_DECLINE_AT(ComputePipelineUnavailable, "no compute pipeline could be created for this dispatch");
return;
}
if (!frame.isCommandRecording) m_frameContext.BeginCommandRecording();
if (!m_uniformManager->BindProgramUniformBuffers(frame.commandBuffer, program, programObj,
m_frameContext.GetCurrentFrameIndex(), VK_PIPELINE_BIND_POINT_COMPUTE)) return;
m_frameContext.GetCurrentFrameIndex(), VK_PIPELINE_BIND_POINT_COMPUTE)) {
MGL_WIRE_DECLINE_AT(ComputeUniformBufferBinding, "the dispatch's uniform/storage bindings could not be bound");
return;
}
vkCmdBindPipeline(frame.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
vkCmdDispatch(frame.commandBuffer, x, y, z);
}
// ---- P8-SV: THE WIRE ARM'S INDIRECT DISPATCH READS ITS GROUP COUNTS ON THE GPU ----------
//
// The monolith arm's shape (VulkanRenderer.cpp's DispatchComputeIndirect) with the wire
// store's own VkBuffer in place of the frontend buffer's resident slice. Before this the wire
// arm read the three words through ReadWireBuffer and dispatched them from the CPU, and a store
// a shader had written made that read a whole-GPU wait (WaitForWireBufferHostAccess) per call.
// There is no shape the device cannot issue, so there is no CPU path left: a record whose
// store or words are unusable is declined by name.
//
// The barrier is P8-D's draw rule (DrawWireIndirectNative): a store marked shader-written gets
// one MEMORY_WRITE -> INDIRECT_COMMAND_READ barrier before its next indirect read, whether or
// not the application issued GL_COMMAND_BARRIER_BIT; DRAW_INDIRECT is the stage that consumes
// VkDispatchIndirectCommand too. It is TAKEN before the bindings below, because binding this
// dispatch's own storage buffers marks them written, and a store that is both this dispatch's
// parameters and one of its outputs must keep that mark for the next reader.
void VulkanRenderer::DispatchWireComputeIndirect(GLintptr offset) {
RetireWireDrawPass();
const auto& state = MG_Pipe::MGPipeApplier();
constexpr Uint64 kCommandSize = 3 * sizeof(Uint32); // GL 4.6 10.3.10 / VkDispatchIndirectCommand
BufferSlice parameters{};
if (MG_Pipe::MGPipeHandleIsNull(state.VerbDispatchIndirectBuffer) ||
!m_bufferManager.AcquireWireSlice(BufferKind::Indirect, state.VerbDispatchIndirectBuffer, parameters)) {
MGL_WIRE_DECLINE_AT(DispatchIndirectBufferUnbound, "an indirect dispatch names no live dispatch-indirect store");
return;
}
// GL refuses an offset that is negative or not a multiple of four (INVALID_VALUE) and a
// command past the end of the buffer (INVALID_OPERATION); a peer's record that breaks
// either is declined, not issued.
if (offset < 0 || (static_cast<Uint64>(offset) & 3u) != 0 || static_cast<Uint64>(offset) > parameters.size ||
parameters.size - static_cast<Uint64>(offset) < kCommandSize) {
MGL_WIRE_DECLINE_AT(DispatchIndirectRange, "the dispatch group counts lie outside their store or are misaligned");
return;
}
auto& frame = m_frameContext.GetCurrent();
if (m_bufferManager.TakeWireIndirectReadBarrier(state.VerbDispatchIndirectBuffer)) {
if (!frame.isCommandRecording) m_frameContext.BeginCommandRecording();
if (VkRenderPassManager::GetActiveRenderPass() != nullptr)
VkRenderPassManager::EndRenderPass(frame.commandBuffer);
VkMemoryBarrier barrier{VK_STRUCTURE_TYPE_MEMORY_BARRIER};
barrier.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT;
vkCmdPipelineBarrier(frame.commandBuffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT, 0, 1, &barrier, 0, nullptr, 0, nullptr);
if (MG_Util::PipeStats::Enabled())
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireIndirectBarriers, 1);
}
const auto program = WireProgramSource(state.DispatchProgram);
if (!program.GetLinkStatus() || !program.GetSpirvStatus()) {
MGL_WIRE_DECLINE_AT(ComputeProgramLinkOrSpirv, "the compute program is not linked or carries no SPIR-V");
return;
}
const auto& programObj = m_programFactory->GetOrCreateProgram(program, {});
if (!m_uniformManager->PrepareWireTextureResources(program, programObj)) {
MGL_WIRE_DECLINE_AT(ComputeTextureResourcesUnprepared, "a sampled/storage image could not be made ready for the dispatch");
return;
}
const auto pipeline = GetOrCreateComputePipeline(programObj);
if (!pipeline) {
MGL_WIRE_DECLINE_AT(ComputePipelineUnavailable, "no compute pipeline could be created for this dispatch");
return;
}
// PrepareWireTextureResources may have flushed and begun a new recording.
auto& dispatchFrame = m_frameContext.GetCurrent();
if (!dispatchFrame.isCommandRecording) m_frameContext.BeginCommandRecording();
if (!m_uniformManager->BindProgramUniformBuffers(dispatchFrame.commandBuffer, program, programObj,
m_frameContext.GetCurrentFrameIndex(), VK_PIPELINE_BIND_POINT_COMPUTE)) {
MGL_WIRE_DECLINE_AT(ComputeUniformBufferBinding, "the dispatch's uniform/storage bindings could not be bound");
return;
}
vkCmdBindPipeline(dispatchFrame.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
vkCmdDispatchIndirect(dispatchFrame.commandBuffer, parameters.buffer,
parameters.offset + static_cast<VkDeviceSize>(offset));
if (MG_Util::PipeStats::Enabled())
MG_Util::PipeStats::AddCalls(MG_Util::PipeStats::CallClass::WireIndirectNativeDispatches, 1);
}
#endif
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,16 @@
#version 450
// SAMPLE ZERO, and it is not an approximation: VK_RESOLVE_MODE_SAMPLE_ZERO_BIT is the mode the
// VK_KHR_depth_stencil_resolve arm asks for, and GL 4.6 core 18.3.1 says a depth resolve picks
// a value "in an implementation-dependent manner where the result will be between the minimum
// and maximum depth values in the pixel" - sample 0 is one of the samples, so it qualifies, and
// taking the same sample as the extension arm is what makes the two arms interchangeable on one
// device.
//
// AVERAGING WOULD NOT BE: a mean of four depths is a surface that exists nowhere, and GL's
// bound is on the per-pixel min/max, which a mean satisfies but a shadow comparison against it
// does not survive.
layout(set = 0, binding = 0) uniform sampler2DMS sourceImage;
void main() {
gl_FragDepth = texelFetch(sourceImage, ivec2(gl_FragCoord.xy), 0).r;
}
@@ -0,0 +1,167 @@
// Included by VulkanRenderer.cpp inside the backend namespace, after WireDepthMipmap.inc.
// The native multisample depth/stencil resolve owns no frontend object and never invokes the
// GL shader compiler.
//
// WHY THIS FILE EXISTS. vkCmdResolveImage is COLOUR ONLY. The one way Vulkan resolves a depth
// or stencil aspect without a shader is a render pass with a depth/stencil resolve attachment,
// which needs VK_KHR_create_renderpass2 and VK_KHR_depth_stencil_resolve - and until this
// package a device without either killed the session at
// `Magma:multisample-depth-resolve-capability` the first time an application resolved a
// multisample depth buffer, which is what glBlitFramebuffer(GL_DEPTH_BUFFER_BIT) from a
// multisampled read framebuffer IS. CONTRACT-P7 §3.2's `multisample-blit-aspect` row names both
// arms; this is the second one.
//
// SAMPLE ZERO IN BOTH ARMS. The extension arm asks for VK_RESOLVE_MODE_SAMPLE_ZERO_BIT, so the
// shader arm fetches sample 0 - not an average. Two arms that answer differently on two devices
// would make "the wire arm is observably the monolith's" untestable, and GL 4.6 core 18.3.1
// requires a SELECTED sample for stencil anyway.
#if MOBILEGL_BUILD_DISAGGREGATED
// The pass's modules, layouts, render passes and pipeline, and the draw it records, are
// WireDepthResolveProbe.cpp's (g5-msprobe): the multisample depth/stencil resolve probe runs
// this exact arm as its CONTROL, so it is defined once for both.
struct VulkanRenderer::WireMultisampleResolveResources {
struct Target {
VkFormat format = VK_FORMAT_UNDEFINED;
VkImageAspectFlags aspect = 0; // the ONE aspect this pipeline writes
VkRenderPass renderPass = VK_NULL_HANDLE;
VkPipeline pipeline = VK_NULL_HANDLE;
};
WireShaderResolveKit kit;
Vector<Target> targets;
};
void VulkanRenderer::DestroyWireMultisampleResolveResources() {
if (!m_wireMultisampleResolveResources) return;
// Shutdown calls this after vkDeviceWaitIdle and after all per-submit descriptor
// pools/framebuffers/views have been collected.
auto& cache = *m_wireMultisampleResolveResources;
for (const auto& target : cache.targets) {
vkDestroyPipeline(m_device, target.pipeline, nullptr);
vkDestroyRenderPass(m_device, target.renderPass, nullptr);
}
DestroyWireShaderResolveKit(m_device, cache.kit);
delete m_wireMultisampleResolveResources;
m_wireMultisampleResolveResources = nullptr;
}
// ONE ASPECT, source -> resolved, both 2D images of the same format and extent, the source
// multisampled and the destination not. Returns false for every shape it cannot express,
// which the caller turns into a decline; it never ends the session.
Bool VulkanRenderer::ResolveWireDepthStencilWithShader(WireImage source, WireImage resolved,
VkImageAspectFlags aspect, GLint sx0, GLint sy0, Uint32 width, Uint32 height) {
if (!source.image || !resolved.image || !width || !height) return false;
if (aspect != VK_IMAGE_ASPECT_DEPTH_BIT && aspect != VK_IMAGE_ASPECT_STENCIL_BIT) return false;
// A fragment shader cannot write the stencil aspect at all without
// VK_EXT_shader_stencil_export. §3.2: decline stencil-only rather than produce a level
// whose stencil is whatever the DONT_CARE left there.
if (aspect == VK_IMAGE_ASPECT_STENCIL_BIT && !m_wireShaderStencilExport) return false;
if (source.samples == VK_SAMPLE_COUNT_1_BIT) return false;
if (source.format != resolved.format) return false;
// SAMPLED is asked for by every attachment VkTextureManager creates; a device that
// declined it would make the descriptor write below undefined, not failed.
if (!(source.usageFlags & VK_IMAGE_USAGE_SAMPLED_BIT)) return false;
VkFormatProperties properties{};
vkGetPhysicalDeviceFormatProperties(m_physicalDevice.handle, source.format, &properties);
const VkFormatFeatureFlags required = VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT |
VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT;
if ((properties.optimalTilingFeatures & required) != required) return false;
const VkImageAspectFlags formatAspects = WireDepthStencilFormatAspects(source.format);
if (!(formatAspects & aspect)) return false;
if (!m_wireMultisampleResolveResources) {
m_wireMultisampleResolveResources = new WireMultisampleResolveResources;
VK_VERIFY(CreateWireShaderResolveKit(m_device, m_wireShaderStencilExport,
m_wireMultisampleResolveResources->kit));
}
auto& cache = *m_wireMultisampleResolveResources;
const VkShaderModule fragment = aspect == VK_IMAGE_ASPECT_STENCIL_BIT ? cache.kit.stencilFragment
: cache.kit.depthFragment;
if (fragment == VK_NULL_HANDLE) return false;
WireMultisampleResolveResources::Target* target = nullptr;
for (auto& entry : cache.targets)
if (entry.format == resolved.format && entry.aspect == aspect) { target = &entry; break; }
if (!target) {
cache.targets.emplace_back();
target = &cache.targets.back();
target->format = resolved.format;
target->aspect = aspect;
VK_VERIFY(CreateWireShaderResolveRenderPass(m_device, resolved.format, &target->renderPass));
if (CreateWireShaderResolvePipeline(m_device, cache.kit, target->renderPass, aspect,
&target->pipeline) != VK_SUCCESS) {
vkDestroyRenderPass(m_device, target->renderPass, nullptr);
cache.targets.pop_back();
return false;
}
}
m_textureManager->FlushPendingUploads();
TransitionWireImage(source, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
TransitionWireImage(resolved, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
// ONE ASPECT on the sampled view: a depth/stencil image can only be SAMPLED through a
// view that names exactly one of its aspects, which is also what makes the stencil half
// a usampler2DMS and the depth half a sampler2DMS.
const VkImageViewCreateInfo sourceViewInfo =
WireDepthStencilViewInfo(source.image, source.format, aspect, source.level, source.layer);
VkImageView sourceView = VK_NULL_HANDLE;
VK_VERIFY(vkCreateImageView(m_device, &sourceViewInfo, nullptr, &sourceView));
// BOTH aspects on the attachment view: a depth/stencil ATTACHMENT names every aspect
// its format has, whichever one the pipeline writes.
const VkImageViewCreateInfo targetViewInfo =
WireDepthStencilViewInfo(resolved.image, resolved.format, formatAspects, 0, 0);
VkImageView targetView = VK_NULL_HANDLE;
VK_VERIFY(vkCreateImageView(m_device, &targetViewInfo, nullptr, &targetView));
VkFramebufferCreateInfo framebufferInfo{VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO};
framebufferInfo.renderPass = target->renderPass;
framebufferInfo.attachmentCount = 1;
framebufferInfo.pAttachments = &targetView;
framebufferInfo.width = resolved.extent.width;
framebufferInfo.height = resolved.extent.height;
framebufferInfo.layers = 1;
VkFramebuffer framebuffer = VK_NULL_HANDLE;
VK_VERIFY(vkCreateFramebuffer(m_device, &framebufferInfo, nullptr, &framebuffer));
// Each in-flight pass owns its descriptor set, for the colour blit's reason: updating a
// cached set would rewrite the source view of an older queued pass.
VkDescriptorPool descriptorPool = VK_NULL_HANDLE;
VkDescriptorSet descriptorSet = VK_NULL_HANDLE;
const VkDescriptorPoolSize poolSize{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1};
VkDescriptorPoolCreateInfo pool{VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO};
pool.maxSets = 1;
pool.poolSizeCount = 1;
pool.pPoolSizes = &poolSize;
VK_VERIFY(vkCreateDescriptorPool(m_device, &pool, nullptr, &descriptorPool));
VkDescriptorSetAllocateInfo alloc{VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO};
alloc.descriptorPool = descriptorPool;
alloc.descriptorSetCount = 1;
alloc.pSetLayouts = &cache.kit.descriptorLayout;
VK_VERIFY(vkAllocateDescriptorSets(m_device, &alloc, &descriptorSet));
const VkDescriptorImageInfo sampled{cache.kit.nearest, sourceView,
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL};
VkWriteDescriptorSet write{VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET};
write.dstSet = descriptorSet;
write.dstBinding = 0;
write.descriptorCount = 1;
write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
write.pImageInfo = &sampled;
vkUpdateDescriptorSets(m_device, 1, &write, 0, nullptr);
auto& frame = m_frameContext.GetCurrent();
RecordWireShaderResolveDraw(frame.commandBuffer, cache.kit, target->renderPass, target->pipeline, framebuffer,
descriptorSet, resolved.extent, {{sx0, sy0}, {width, height}});
ResetDynamicStateShadow();
InvalidatePipelineMemo();
// This native pass binds its own set/layout without going through the application's
// descriptor binder. Its next fast path must rebind.
if (m_uniformManager) m_uniformManager->OnCommandBufferBoundary();
WireRetiredObjects retired{};
retired.framebuffers.push_back(framebuffer);
retired.imageViews = {sourceView, targetView};
retired.descriptorPools.push_back(descriptorPool);
RetireWireObjects(Move(retired));
return true;
}
#endif
@@ -0,0 +1,18 @@
#version 450
// Native Vulkan multisample depth/stencil resolve: no GL program, registry entry, or runtime
// transpilation.
//
// THIS STAGE CARRIES NOTHING, and that is the whole design. The scratch this pass writes has
// the SAME extent and the SAME coordinate space as the multisample source it reads - both are
// native images, neither is flipped against the other - so the fragment stage can take its
// coordinate straight out of gl_FragCoord and needs no varying, no push constant and no rect.
// The requested rectangle is expressed by the scissor instead, which is also what the
// VK_KHR_depth_stencil_resolve arm next door expresses with its renderArea.
//
// An oversized triangle rather than WireDepthMipmap.vert's quad: there is no destination
// rectangle to clip to here, only the scissor, and a triangle rasterizes the same area with
// one fewer primitive and no shared edge.
void main() {
const vec2 corners[3] = vec2[](vec2(-1.0, -1.0), vec2(3.0, -1.0), vec2(-1.0, 3.0));
gl_Position = vec4(corners[gl_VertexIndex], 0.0, 1.0);
}
@@ -0,0 +1,103 @@
// Generated from WireMultisampleResolve.vert / WireMultisampleDepthResolve.frag /
// WireMultisampleStencilResolve.frag by scripts/bake_internal_shaders.py, which runs MG_Test's
// BakedInternalShadersTest with the IN-TREE glslang and splices the words it emits into the
// arrays below. Do not edit the words by hand: MOBILEGL_BAKED_INTERNAL_SHADERS (the same test,
// run as a gate) recompiles every source in its table and reds on the first word that
// disagrees.
#pragma once
static constexpr Uint32 kWireMultisampleResolveVertexSpirv[] = {
0x07230203u, 0x00010000u, 0x0008000bu, 0x00000028u, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
0x0007000fu, 0x00000000u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x0000000du, 0x0000001au, 0x00030003u,
0x00000002u, 0x000001c2u, 0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00060005u, 0x0000000bu,
0x505f6c67u, 0x65567265u, 0x78657472u, 0x00000000u, 0x00060006u, 0x0000000bu, 0x00000000u, 0x505f6c67u,
0x7469736fu, 0x006e6f69u, 0x00070006u, 0x0000000bu, 0x00000001u, 0x505f6c67u, 0x746e696fu, 0x657a6953u,
0x00000000u, 0x00070006u, 0x0000000bu, 0x00000002u, 0x435f6c67u, 0x4470696cu, 0x61747369u, 0x0065636eu,
0x00070006u, 0x0000000bu, 0x00000003u, 0x435f6c67u, 0x446c6c75u, 0x61747369u, 0x0065636eu, 0x00030005u,
0x0000000du, 0x00000000u, 0x00060005u, 0x0000001au, 0x565f6c67u, 0x65747265u, 0x646e4978u, 0x00007865u,
0x00050005u, 0x0000001du, 0x65646e69u, 0x6c626178u, 0x00000065u, 0x00030047u, 0x0000000bu, 0x00000002u,
0x00050048u, 0x0000000bu, 0x00000000u, 0x0000000bu, 0x00000000u, 0x00050048u, 0x0000000bu, 0x00000001u,
0x0000000bu, 0x00000001u, 0x00050048u, 0x0000000bu, 0x00000002u, 0x0000000bu, 0x00000003u, 0x00050048u,
0x0000000bu, 0x00000003u, 0x0000000bu, 0x00000004u, 0x00040047u, 0x0000001au, 0x0000000bu, 0x0000002au,
0x00020013u, 0x00000002u, 0x00030021u, 0x00000003u, 0x00000002u, 0x00030016u, 0x00000006u, 0x00000020u,
0x00040017u, 0x00000007u, 0x00000006u, 0x00000004u, 0x00040015u, 0x00000008u, 0x00000020u, 0x00000000u,
0x0004002bu, 0x00000008u, 0x00000009u, 0x00000001u, 0x0004001cu, 0x0000000au, 0x00000006u, 0x00000009u,
0x0006001eu, 0x0000000bu, 0x00000007u, 0x00000006u, 0x0000000au, 0x0000000au, 0x00040020u, 0x0000000cu,
0x00000003u, 0x0000000bu, 0x0004003bu, 0x0000000cu, 0x0000000du, 0x00000003u, 0x00040015u, 0x0000000eu,
0x00000020u, 0x00000001u, 0x0004002bu, 0x0000000eu, 0x0000000fu, 0x00000000u, 0x00040017u, 0x00000010u,
0x00000006u, 0x00000002u, 0x0004002bu, 0x00000008u, 0x00000011u, 0x00000003u, 0x0004001cu, 0x00000012u,
0x00000010u, 0x00000011u, 0x0004002bu, 0x00000006u, 0x00000013u, 0xbf800000u, 0x0005002cu, 0x00000010u,
0x00000014u, 0x00000013u, 0x00000013u, 0x0004002bu, 0x00000006u, 0x00000015u, 0x40400000u, 0x0005002cu,
0x00000010u, 0x00000016u, 0x00000015u, 0x00000013u, 0x0005002cu, 0x00000010u, 0x00000017u, 0x00000013u,
0x00000015u, 0x0006002cu, 0x00000012u, 0x00000018u, 0x00000014u, 0x00000016u, 0x00000017u, 0x00040020u,
0x00000019u, 0x00000001u, 0x0000000eu, 0x0004003bu, 0x00000019u, 0x0000001au, 0x00000001u, 0x00040020u,
0x0000001cu, 0x00000007u, 0x00000012u, 0x00040020u, 0x0000001eu, 0x00000007u, 0x00000010u, 0x0004002bu,
0x00000006u, 0x00000021u, 0x00000000u, 0x0004002bu, 0x00000006u, 0x00000022u, 0x3f800000u, 0x00040020u,
0x00000026u, 0x00000003u, 0x00000007u, 0x00050036u, 0x00000002u, 0x00000004u, 0x00000000u, 0x00000003u,
0x000200f8u, 0x00000005u, 0x0004003bu, 0x0000001cu, 0x0000001du, 0x00000007u, 0x0004003du, 0x0000000eu,
0x0000001bu, 0x0000001au, 0x0003003eu, 0x0000001du, 0x00000018u, 0x00050041u, 0x0000001eu, 0x0000001fu,
0x0000001du, 0x0000001bu, 0x0004003du, 0x00000010u, 0x00000020u, 0x0000001fu, 0x00050051u, 0x00000006u,
0x00000023u, 0x00000020u, 0x00000000u, 0x00050051u, 0x00000006u, 0x00000024u, 0x00000020u, 0x00000001u,
0x00070050u, 0x00000007u, 0x00000025u, 0x00000023u, 0x00000024u, 0x00000021u, 0x00000022u, 0x00050041u,
0x00000026u, 0x00000027u, 0x0000000du, 0x0000000fu, 0x0003003eu, 0x00000027u, 0x00000025u, 0x000100fdu,
0x00010038u,
};
static constexpr Uint32 kWireMultisampleDepthResolveFragmentSpirv[] = {
0x07230203u, 0x00010000u, 0x0008000bu, 0x0000001du, 0x00000000u, 0x00020011u, 0x00000001u, 0x0006000bu,
0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u, 0x0003000eu, 0x00000000u, 0x00000001u,
0x0007000fu, 0x00000004u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00000008u, 0x00000010u, 0x00030010u,
0x00000004u, 0x00000007u, 0x00030010u, 0x00000004u, 0x0000000cu, 0x00030003u, 0x00000002u, 0x000001c2u,
0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u, 0x00060005u, 0x00000008u, 0x465f6c67u, 0x44676172u,
0x68747065u, 0x00000000u, 0x00050005u, 0x0000000cu, 0x72756f73u, 0x6d496563u, 0x00656761u, 0x00060005u,
0x00000010u, 0x465f6c67u, 0x43676172u, 0x64726f6fu, 0x00000000u, 0x00040047u, 0x00000008u, 0x0000000bu,
0x00000016u, 0x00040047u, 0x0000000cu, 0x00000021u, 0x00000000u, 0x00040047u, 0x0000000cu, 0x00000022u,
0x00000000u, 0x00040047u, 0x00000010u, 0x0000000bu, 0x0000000fu, 0x00020013u, 0x00000002u, 0x00030021u,
0x00000003u, 0x00000002u, 0x00030016u, 0x00000006u, 0x00000020u, 0x00040020u, 0x00000007u, 0x00000003u,
0x00000006u, 0x0004003bu, 0x00000007u, 0x00000008u, 0x00000003u, 0x00090019u, 0x00000009u, 0x00000006u,
0x00000001u, 0x00000000u, 0x00000000u, 0x00000001u, 0x00000001u, 0x00000000u, 0x0003001bu, 0x0000000au,
0x00000009u, 0x00040020u, 0x0000000bu, 0x00000000u, 0x0000000au, 0x0004003bu, 0x0000000bu, 0x0000000cu,
0x00000000u, 0x00040017u, 0x0000000eu, 0x00000006u, 0x00000004u, 0x00040020u, 0x0000000fu, 0x00000001u,
0x0000000eu, 0x0004003bu, 0x0000000fu, 0x00000010u, 0x00000001u, 0x00040017u, 0x00000011u, 0x00000006u,
0x00000002u, 0x00040015u, 0x00000014u, 0x00000020u, 0x00000001u, 0x00040017u, 0x00000015u, 0x00000014u,
0x00000002u, 0x0004002bu, 0x00000014u, 0x00000017u, 0x00000000u, 0x00040015u, 0x0000001au, 0x00000020u,
0x00000000u, 0x0004002bu, 0x0000001au, 0x0000001bu, 0x00000000u, 0x00050036u, 0x00000002u, 0x00000004u,
0x00000000u, 0x00000003u, 0x000200f8u, 0x00000005u, 0x0004003du, 0x0000000au, 0x0000000du, 0x0000000cu,
0x0004003du, 0x0000000eu, 0x00000012u, 0x00000010u, 0x0007004fu, 0x00000011u, 0x00000013u, 0x00000012u,
0x00000012u, 0x00000000u, 0x00000001u, 0x0004006eu, 0x00000015u, 0x00000016u, 0x00000013u, 0x00040064u,
0x00000009u, 0x00000018u, 0x0000000du, 0x0007005fu, 0x0000000eu, 0x00000019u, 0x00000018u, 0x00000016u,
0x00000040u, 0x00000017u, 0x00050051u, 0x00000006u, 0x0000001cu, 0x00000019u, 0x00000000u, 0x0003003eu,
0x00000008u, 0x0000001cu, 0x000100fdu, 0x00010038u,
};
static constexpr Uint32 kWireMultisampleStencilResolveFragmentSpirv[] = {
0x07230203u, 0x00010000u, 0x0008000bu, 0x0000001fu, 0x00000000u, 0x00020011u, 0x00000001u, 0x00020011u,
0x00001395u, 0x0009000au, 0x5f565053u, 0x5f545845u, 0x64616873u, 0x735f7265u, 0x636e6574u, 0x655f6c69u,
0x726f7078u, 0x00000074u, 0x0006000bu, 0x00000001u, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0x00000000u,
0x0003000eu, 0x00000000u, 0x00000001u, 0x0007000fu, 0x00000004u, 0x00000004u, 0x6e69616du, 0x00000000u,
0x00000008u, 0x00000012u, 0x00030010u, 0x00000004u, 0x00000007u, 0x00030010u, 0x00000004u, 0x000013a3u,
0x00030003u, 0x00000002u, 0x000001c2u, 0x00090004u, 0x415f4c47u, 0x735f4252u, 0x65646168u, 0x74735f72u,
0x69636e65u, 0x78655f6cu, 0x74726f70u, 0x00000000u, 0x00040005u, 0x00000004u, 0x6e69616du, 0x00000000u,
0x00080005u, 0x00000008u, 0x465f6c67u, 0x53676172u, 0x636e6574u, 0x65526c69u, 0x42524166u, 0x00000000u,
0x00050005u, 0x0000000du, 0x72756f73u, 0x6d496563u, 0x00656761u, 0x00060005u, 0x00000012u, 0x465f6c67u,
0x43676172u, 0x64726f6fu, 0x00000000u, 0x00040047u, 0x00000008u, 0x0000000bu, 0x00001396u, 0x00040047u,
0x0000000du, 0x00000021u, 0x00000000u, 0x00040047u, 0x0000000du, 0x00000022u, 0x00000000u, 0x00040047u,
0x00000012u, 0x0000000bu, 0x0000000fu, 0x00020013u, 0x00000002u, 0x00030021u, 0x00000003u, 0x00000002u,
0x00040015u, 0x00000006u, 0x00000020u, 0x00000001u, 0x00040020u, 0x00000007u, 0x00000003u, 0x00000006u,
0x0004003bu, 0x00000007u, 0x00000008u, 0x00000003u, 0x00040015u, 0x00000009u, 0x00000020u, 0x00000000u,
0x00090019u, 0x0000000au, 0x00000009u, 0x00000001u, 0x00000000u, 0x00000000u, 0x00000001u, 0x00000001u,
0x00000000u, 0x0003001bu, 0x0000000bu, 0x0000000au, 0x00040020u, 0x0000000cu, 0x00000000u, 0x0000000bu,
0x0004003bu, 0x0000000cu, 0x0000000du, 0x00000000u, 0x00030016u, 0x0000000fu, 0x00000020u, 0x00040017u,
0x00000010u, 0x0000000fu, 0x00000004u, 0x00040020u, 0x00000011u, 0x00000001u, 0x00000010u, 0x0004003bu,
0x00000011u, 0x00000012u, 0x00000001u, 0x00040017u, 0x00000013u, 0x0000000fu, 0x00000002u, 0x00040017u,
0x00000016u, 0x00000006u, 0x00000002u, 0x0004002bu, 0x00000006u, 0x00000018u, 0x00000000u, 0x00040017u,
0x0000001au, 0x00000009u, 0x00000004u, 0x0004002bu, 0x00000009u, 0x0000001cu, 0x00000000u, 0x00050036u,
0x00000002u, 0x00000004u, 0x00000000u, 0x00000003u, 0x000200f8u, 0x00000005u, 0x0004003du, 0x0000000bu,
0x0000000eu, 0x0000000du, 0x0004003du, 0x00000010u, 0x00000014u, 0x00000012u, 0x0007004fu, 0x00000013u,
0x00000015u, 0x00000014u, 0x00000014u, 0x00000000u, 0x00000001u, 0x0004006eu, 0x00000016u, 0x00000017u,
0x00000015u, 0x00040064u, 0x0000000au, 0x00000019u, 0x0000000eu, 0x0007005fu, 0x0000001au, 0x0000001bu,
0x00000019u, 0x00000017u, 0x00000040u, 0x00000018u, 0x00050051u, 0x00000009u, 0x0000001du, 0x0000001bu,
0x00000000u, 0x0004007cu, 0x00000006u, 0x0000001eu, 0x0000001du, 0x0003003eu, 0x00000008u, 0x0000001eu,
0x000100fdu, 0x00010038u,
};
@@ -0,0 +1,19 @@
#version 450
// The stencil half, and it needs an extension to exist at all: a fragment shader cannot write
// the stencil aspect without VK_EXT_shader_stencil_export (SPV_EXT_shader_stencil_export, which
// this `#extension` line is how glslang is asked for). CONTRACT-P7 §3.2 says so and says what
// happens otherwise - a stencil-only resolve DECLINES on a device without it, because there is
// no second way to put a value in the stencil aspect from a pass.
//
// usampler2DMS and not sampler2DMS: the stencil aspect of a depth/stencil image is an unsigned
// integer, and a float view of it would round the index.
//
// Sample zero, for WireMultisampleDepthResolve.frag's reason, and GL is stricter here: 18.3.1
// says integer and stencil resolves SELECT one sample rather than combining them, so sample 0
// is the answer and not merely an acceptable one.
#extension GL_ARB_shader_stencil_export : require
layout(set = 0, binding = 0) uniform usampler2DMS sourceImage;
void main() {
gl_FragStencilRefARB = int(texelFetch(sourceImage, ivec2(gl_FragCoord.xy), 0).r);
}
@@ -13,9 +13,28 @@
image = {};
}
void UniformManager::SetWireInvalidStorageImageArm(Bool deviceNullDescriptor) {
const char* forced = std::getenv("MGITEST_MAGMA_FORCE_PRIVATE_IMAGE_PLACEHOLDER");
const Bool forcePrivate = forced != nullptr && std::strcmp(forced, "1") == 0;
m_wireInvalidStorageImagesBindNull = deviceNullDescriptor && !forcePrivate;
MGLOG_I("Magma wire: an invalid or unbound image unit binds %s (VK_EXT_robustness2 nullDescriptor %s%s)",
m_wireInvalidStorageImagesBindNull ? "a NULL storage descriptor"
: "a placeholder private to its unit",
deviceNullDescriptor ? "enabled" : "unavailable",
deviceNullDescriptor && forcePrivate ? "; MGITEST_MAGMA_FORCE_PRIVATE_IMAGE_PLACEHOLDER=1" : "");
}
Bool UniformManager::ResolveWirePlaceholderImage(VkCommandBuffer commands, const MagmaProgramSource& program,
const ProgramFactory::VkProgramObject& programObj, Uint32 binding, Bool storage,
VkDescriptorImageInfo& out) const {
VkDescriptorImageInfo& out, VkFormat boundStorageFormat, Uint32 unit) const {
// GL 4.6 core 8.26, codex closeout finding 2: every access through an invalid image unit
// is invalid - a load returns zero, a store and an atomic are discarded. A NULL storage
// descriptor (VK_EXT_robustness2 nullDescriptor) is exactly that, for any target, format
// and sample count, so nothing below is needed.
if (storage && m_wireInvalidStorageImagesBindNull) {
out = {VK_NULL_HANDLE, VK_NULL_HANDLE, VK_IMAGE_LAYOUT_GENERAL};
return true;
}
if (binding >= programObj.samplerTextureTargetByBinding.size() ||
binding >= programObj.samplerNumericDomainByBinding.size()) return false;
const auto target = programObj.samplerTextureTargetByBinding[binding];
@@ -36,6 +55,20 @@
if (storage) {
if (binding >= programObj.storageImageFormatByBinding.size()) return false;
format = programObj.storageImageFormatByBinding[binding];
// A formatless declaration of an INVALID binding (a unit whose texture lacks the
// named level or layer) takes the format the application bound, where the device can
// store to it and it is in the shader's numeric class. The R32 default expands a load
// to (0, 0, 0, 1); a four-channel carrier cleared to zero reads (0, 0, 0, 0), which
// is what KHR-GL46.shader_image_load_store.incomplete_textures compares against.
if (format == VK_FORMAT_UNDEFINED && boundStorageFormat != VK_FORMAT_UNDEFINED) {
const Bool classMatches = domain == SamplerNumericDomain::Float
? !vkuFormatIsUINT(boundStorageFormat) && !vkuFormatIsSINT(boundStorageFormat)
: domain == SamplerNumericDomain::SignedInteger ? vkuFormatIsSINT(boundStorageFormat)
: domain == SamplerNumericDomain::UnsignedInteger && vkuFormatIsUINT(boundStorageFormat);
VkFormatProperties properties{};
if (classMatches) vkGetPhysicalDeviceFormatProperties(m_physicalDevice, boundStorageFormat, &properties);
if (properties.optimalTilingFeatures & VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT) format = boundStorageFormat;
}
if (format == VK_FORMAT_UNDEFINED) format = PlaceholderFormatForNumericDomain(domain);
} else if (shadow) format = VK_FORMAT_D16_UNORM;
else if (domain == SamplerNumericDomain::Float) format = VK_FORMAT_R8G8B8A8_UNORM;
@@ -60,8 +93,22 @@
break;
default: return false;
}
const Uint64 key = static_cast<Uint32>(format) | (static_cast<Uint64>(target) << 32) |
(static_cast<Uint64>(storage) << 48) | (static_cast<Uint64>(shadow) << 49);
// A STORAGE placeholder is private to its unit (per shape). Shared by shape alone, a store
// through invalid unit A was visible, after a barrier, to a load through invalid unit B of
// the same shape - one Vulkan image behind both (codex closeout finding 2). The clear
// below still runs before each use; what a private image cannot hide is the SAME unit's
// own store to its own later load inside one pass (CONTRACT-P7 §12). Not per binding: GL
// aliases per unit, and a (binding, unit) key only multiplied allocations. Past
// kWirePrivateStoragePlaceholderCap a new (shape, unit) shares its shape's (WirePlaceholderKey.h).
const Uint64 shape = static_cast<Uint32>(format) | (static_cast<Uint64>(target) << 32) |
(static_cast<Uint64>(storage) << 48) | (static_cast<Uint64>(shadow) << 49);
const WirePlaceholderKey key = ChooseWirePlaceholderKey(shape, storage, unit, m_wirePrivateStoragePlaceholders,
[this](const WirePlaceholderKey& candidate) { return m_wirePlaceholderImages.count(candidate) != 0; });
if (storage && unit != WirePlaceholderKey::kShared && key.unit == WirePlaceholderKey::kShared) {
MGLOG_W_ONCE("Magma wire: %zu image-unit-private placeholders exist (the cap); a further invalid or "
"unbound unit shares its shape's placeholder, so two such units of one shape can alias",
static_cast<size_t>(m_wirePrivateStoragePlaceholders));
}
auto found = m_wirePlaceholderImages.find(key);
if (found == m_wirePlaceholderImages.end()) {
WirePlaceholderImage image{};
@@ -151,11 +198,12 @@
if (vkCreateFramebuffer(m_device, &framebuffer, nullptr, &image.clearFramebuffer) != VK_SUCCESS) return fail();
}
found = m_wirePlaceholderImages.emplace(key, image).first;
if (key.unit != WirePlaceholderKey::kShared) ++m_wirePrivateStoragePlaceholders;
}
auto& image = found->second;
// Sampled incomplete textures remain immutable opaque black. A storage
// placeholder is reset before each use, so a prior discarded store cannot
// leak into an unrelated invocation's unbound image loads.
// leak into a later pass's loads through the same unit.
if (storage || image.layout == VK_IMAGE_LAYOUT_UNDEFINED) {
VkClearValue clear{};
if (!storage) {
@@ -0,0 +1,62 @@
// MobileGL - MobileGL/MG_Backend/DirectVulkan/Renderer/WirePlaceholderKey.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Includes.h>
// The pure half of UniformManager's wire placeholder cache (WirePlaceholderImages.inc): WHICH
// placeholder a binding with no usable image takes. A unit test pins it
// (MG_Test/Pipeline/PipelineQuirkTest.cpp); the Vulkan half that makes the image stays in the .inc.
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Pipe/MGPipeTypes.h>
namespace MobileGL::MG_Backend::DirectVulkan {
// The placeholder's SHAPE (format, target, storage, shadow) and, for a STORAGE placeholder on
// the arm without a null descriptor, the image UNIT it is private to. GL aliases per UNIT, so the
// unit is the whole of the no-alias guarantee (codex closeout finding 2): two bindings that name
// one unit are one GL image, and keying them apart as well only multiplied allocations (cf-magma
// critic). A sampled placeholder is never written, so it stays shared: its unit is kShared.
struct WirePlaceholderKey {
static constexpr Uint32 kShared = ~0u;
Uint64 shape = 0;
Uint32 unit = kShared;
Bool operator==(const WirePlaceholderKey& other) const { return shape == other.shape && unit == other.unit; }
};
struct WirePlaceholderKeyHash {
SizeT operator()(const WirePlaceholderKey& key) const {
Uint64 h = key.shape * 0x9E3779B97F4A7C15ull;
h ^= static_cast<Uint64>(key.unit) + 0x9E3779B97F4A7C15ull + (h << 6) + (h >> 2);
return static_cast<SizeT>(h);
}
};
// Every private placeholder is its own vkCreateImage + vkAllocateMemory, freed only at Shutdown.
// (shape, unit) bounds them by shapes x kMGPipeMaxImageUnits, and a peer rotating formats and
// targets across the units could still walk that towards maxMemoryAllocationCount (4096 is the
// spec minimum). So past this many, a NEW (shape, unit) takes its shape's SHARED placeholder - the
// pre-fix aliasing, logged once by the caller - instead of another allocation; a unit that already
// has its private placeholder keeps it. Two per unit is far past any real program.
inline constexpr SizeT kWirePrivateStoragePlaceholderCap = 2 * static_cast<SizeT>(MG_Pipe::kMGPipeMaxImageUnits);
// The key a placeholder of `shape` is found or made under. `privateStorage` = a storage binding on
// the arm without a null descriptor (with one, the caller never gets here); `unit` = its image
// unit; `privateCount` = how many private placeholders exist; `known(key)` = the cache holds `key`.
template <typename KnownFn>
WirePlaceholderKey ChooseWirePlaceholderKey(Uint64 shape, Bool privateStorage, Uint32 unit, SizeT privateCount,
KnownFn&& known) {
WirePlaceholderKey key;
key.shape = shape;
if (!privateStorage || unit == WirePlaceholderKey::kShared) return key;
key.unit = unit;
if (privateCount < kWirePrivateStoragePlaceholderCap || known(key)) return key;
key.unit = WirePlaceholderKey::kShared;
return key;
}
} // namespace MobileGL::MG_Backend::DirectVulkan
#endif
@@ -22,6 +22,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
Uint64 CacheVersion = MG_Config::CacheVersion;
Uint32 SurfaceWidth = 1;
Uint32 SurfaceHeight = 1;
// The app's eglSwapInterval; empty until it asks. Window renderers only - see
// VulkanRenderer::SetSwapInterval.
Optional<Int> SwapInterval;
Bool DisablePipelineCache = false;
#if MOBILEGL_LOG_ACTIVE_LEVEL <= MOBILEGL_LOG_LEVEL_DEBUG
Bool EnableValidationLayers = true;
+737
View File
@@ -0,0 +1,737 @@
// MobileGL - MobileGL/MG_Backend/GbmFrameChannel.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "GbmFrameChannel.h"
#if MOBILEGL_BUILD_DISAGGREGATED && defined(__ANDROID__)
#include <android/hardware_buffer.h>
#include <cerrno>
#include <cstring>
#include <poll.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/un.h>
#include <unistd.h>
// WHAT THIS FILE IS WRITTEN AGAINST, AND WHAT HAS NOT BEEN RUN.
//
// The other end exists and is tested on this machine: MobileGL/MG_Gbm/GbmHostChannel.cpp dials
// this socket with SOCK_SEQPACKET, reads one message per frame, computes that message's expected
// length from the handle's own two counts, and refuses any other length. Every byte below is
// written to that parser. There is no phone attached to the machine this was written on, so the
// file is compile-verified for aarch64 Android and nothing more: the first run on a device is what
// will say whether the relay in RelayHandle() reads back the handle's bytes exactly as
// AHardwareBuffer_sendHandleToUnixSocket writes them, which is why that path is written to
// survive both layouts the platform's serializer could have (see the note there).
namespace MobileGL::MG_Backend {
namespace {
// THE LAYOUT IS A PROTOCOL, SO ITS SIZES ARE ASSERTED AND NOT ASSUMED. The container
// derives the length of a frame message from these three sizes plus the handle's counts,
// so a field added here, or the packing attribute dropped, would not be a compile error
// over there - it would be every frame skipped with a message about its length.
static_assert(sizeof(GbmFrameOffer) == 10 * sizeof(Uint32), "the offer is ten 32 bit fields on the wire");
static_assert(sizeof(GbmFrameRelease) == 3 * sizeof(Uint32), "the release is three 32 bit fields on the wire");
static_assert(sizeof(GbmNativeHandleHeader) == 3 * sizeof(Int32),
"a native handle begins with a version and two counts, on every Android ABI");
// The container reads one message per frame into a 4 KiB buffer and skips a message that
// does not fit in it (GbmHostMessageMax, GbmHostChannel.cpp), so a message larger than this
// is refused here rather than built and dropped there.
constexpr SizeT kMessageMax = 4096;
// The descriptors one handle may carry, which is the container's own cap (GbmHostMaxFds).
// Every AHardwareBuffer handle this project has seen carries one - the dma-buf - and both
// ends refuse a frame whose handle names more than they can take rather than forward a
// truncated list into a buffer of unknown memory.
constexpr Int kMaxHandleFds = 8;
// How many release messages one call reads before it returns to the frame. A cap and not a
// loop to exhaustion: the messages left behind stay in the socket, whose buffer the kernel
// bounds, and the next frame drains them.
constexpr Int kDrainMax = 64;
// HOW MANY DIALERS THE KERNEL MAY HOLD FOR THIS LISTENER. One container is the protocol,
// but the queue is not the connection: every compositor that was restarted dials once more,
// and a listener that only ever has one slot leaves its queue full for the next one. With
// listen(fd, 1) that is exactly what was measured - connect() from the container answered
// EAGAIN(11) on four attempts out of four after two restarts, and no dialer reached this
// side again for as long as the server lived.
constexpr Int kDialerBacklog = 16;
// The acceptor's poll timeout. It is a wake-up and not a schedule, so it only has to be
// short enough that a container dialling into a socket nothing else touches is taken
// promptly; a dialer that arrives while the thread is waiting wakes it at once.
constexpr Int kAcceptPollMs = 100;
// How long the relay read waits for bytes the send one statement earlier has already put in
// the socket. It exists only so that an invariant broken somewhere else cannot park the
// apply thread; on a local socketpair the wait is zero.
constexpr Int kRelayWaitMs = 2;
// close() may overwrite the errno a caller is in the middle of reporting, and every caller
// here is reporting a failure errno describes.
void CloseQuietly(Int fd) {
if (fd < 0) return;
const Int savedErrno = errno;
::close(fd);
errno = savedErrno;
}
}
GbmFrameChannel::~GbmFrameChannel() { Close(); }
Bool GbmFrameChannel::Open() {
const std::lock_guard<std::mutex> lock(m_mutex);
if (m_listen >= 0) return true;
const char* configured = ::getenv(kGbmFrameSocketEnv);
const char* path = (configured != nullptr && configured[0] != 0) ? configured : kGbmFrameSocketPath;
constexpr SizeT pathMax = sizeof(((struct sockaddr_un*)nullptr)->sun_path);
if (std::strlen(path) >= pathMax) {
MGLOG_E_ONCE("gbm frame channel: the container socket path %s does not fit in sun_path (%zu bytes); no "
"container can be served",
path, pathMax);
return false;
}
// SOCK_SEQPACKET because that is what the container dials with (GbmHostChannel::EnsureOpen
// asks for exactly this), and a stream listener answers it EPROTOTYPE - the container
// prints that as "a socket of a different type". SOCK_NONBLOCK because the accept below
// runs on the apply thread and may not wait for a container that is not there; SOCK_CLOEXEC
// because every process this server starts would otherwise inherit a socket that receives
// the phone's frames.
const Int fd = ::socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0);
if (fd < 0) {
MGLOG_E_ONCE("gbm frame channel: socket(AF_UNIX, SOCK_SEQPACKET): %s; the frames are drawn and presented "
"without a container",
std::strerror(errno));
return false;
}
// A node left behind by a server that died is not a reason to refuse a container: bind
// would answer EADDRINUSE for a socket nobody is listening on.
::unlink(path);
struct sockaddr_un address = {};
address.sun_family = AF_UNIX;
// The terminator travels with the name: sun_path is a C string for a pathname socket, and a
// name without one is a name the kernel reads past the end of.
std::memcpy(address.sun_path, path, std::strlen(path) + 1);
if (::bind(fd, reinterpret_cast<struct sockaddr*>(&address), sizeof(address)) != 0) {
MGLOG_E_ONCE("gbm frame channel: bind(%s): %s; the frames are drawn and presented without a container",
path, std::strerror(errno));
CloseQuietly(fd);
return false;
}
// connect() to a pathname socket needs write permission on the node, and the container is
// not this process's user, so the umask's usual 0755 would leave it refused for a reason it
// could not see. 0666 is what makes the node reachable by the other half of a split run; a
// chmod that fails is not fatal, because a peer running as this uid still dials it.
if (::chmod(path, 0666) != 0) {
MGLOG_W_ONCE("gbm frame channel: chmod(%s, 0666): %s; a container running as another user cannot dial "
"this socket",
path, std::strerror(errno));
}
if (::listen(fd, kDialerBacklog) != 0) {
MGLOG_E_ONCE("gbm frame channel: listen(%s): %s; the frames are drawn and presented without a container",
path, std::strerror(errno));
::unlink(path);
CloseQuietly(fd);
return false;
}
m_listen = fd;
m_path = path;
// THE ACCEPTOR IS STARTED WITH THE SOCKET AND NOT WITH THE FIRST FRAME. A compositor dials
// at its device creation and keeps the connection, so the dial a frame would have answered
// has already happened by the time that frame is offered: an accept on the offer path is an
// accept that runs after the thing it exists for. The descriptor is copied into the thread
// as an argument - it is the one piece of this object the thread reads without the lock,
// which is what lets Close() join it before closing that descriptor.
m_accepting.store(true, std::memory_order_release);
try {
m_acceptor = std::thread(&GbmFrameChannel::AcceptLoop, this, fd);
} catch (...) {
// A thread that cannot be made is a phone that is out of them, not a reason to refuse
// the frames the container can still be given one at a time.
m_accepting.store(false, std::memory_order_release);
MGLOG_W_ONCE("gbm frame channel: the acceptor thread could not be started; a container is only taken in "
"when a frame is offered, and one that dials between frames waits for the next one");
}
MGLOG_I("gbm frame channel: listening on %s for one container (SOCK_SEQPACKET, one message per frame)", path);
return true;
}
Bool GbmFrameChannel::EnsureRelay() {
if (m_relay[0] >= 0 && m_relay[1] >= 0) return true;
// Both ends are dropped before a new pair is made: half a socketpair is a socket with bytes
// in it that nothing will ever read, and the next handle read from it would be the tail of
// the last one.
DropRelay();
Int pair[2] = {-1, -1};
if (::socketpair(AF_UNIX, SOCK_STREAM | SOCK_NONBLOCK | SOCK_CLOEXEC, 0, pair) != 0) {
MGLOG_E_ONCE("gbm frame channel: socketpair for the handle relay: %s; no frame can be offered until one "
"can be made",
std::strerror(errno));
return false;
}
m_relay[0] = pair[0];
m_relay[1] = pair[1];
return true;
}
void GbmFrameChannel::DropRelay() {
CloseQuietly(m_relay[0]);
CloseQuietly(m_relay[1]);
m_relay[0] = -1;
m_relay[1] = -1;
}
Bool GbmFrameChannel::RelayHandle(struct AHardwareBuffer* buffer, Uint8* payload, SizeT capacity, SizeT& payloadBytes,
Int* fds, Int& fdCount) {
payloadBytes = 0;
fdCount = 0;
if (!EnsureRelay()) return false;
// WHY A FRAME'S HANDLE GOES THROUGH A SOCKETPAIR OF THIS CHANNEL'S OWN.
//
// The container takes ONE message per frame - the offer, then the handle's counts, then the
// arrays those counts describe, with the descriptors in that same message's SCM_RIGHTS
// control data - and it refuses a message of any other length. Two sends cannot be one
// message on a message-oriented socket, so AHardwareBuffer_sendHandleToUnixSocket, which
// sends a handle as a message of its own, is pointed at a socketpair this channel owns, and
// what comes out of the other end is both halves of what the container wants: the handle's
// own bytes, read here, and its descriptors, as SCM_RIGHTS.
//
// The alternative does not exist: the NDK's <android/hardware_buffer.h> declares no way to
// ask a buffer for its descriptors. AHardwareBuffer_getNativeHandle is a platform API and
// is absent from the NDK's copy of that header (checked against NDK 27.3.13750724 in this
// tree's build-termux toolchain), so this is the one route from an AHardwareBuffer to a
// dma-buf descriptor that an app can take.
const Int code = AHardwareBuffer_sendHandleToUnixSocket(buffer, m_relay[0]);
if (code != 0) {
MGLOG_W_ONCE("gbm frame channel: the frame's handle did not cross the relay socket (rc %d, errno %d); the "
"frame is not offered to the container",
code, errno);
// The relay is dropped rather than kept: a send that failed part way through would
// leave bytes that the next handle's read would take for its own header.
DropRelay();
return false;
}
struct iovec vector = {};
vector.iov_base = payload;
vector.iov_len = capacity;
Uint8 control[CMSG_SPACE(sizeof(Int) * (SizeT)kMaxHandleFds)] = {};
struct msghdr message = {};
message.msg_iov = &vector;
message.msg_iovlen = 1;
message.msg_control = control;
message.msg_controllen = sizeof(control);
// MSG_DONTWAIT for the promise at the top of this file, and MSG_CMSG_CLOEXEC so a descriptor
// read out of the relay does not survive into a child of this process, which would then hold
// the phone's frame until it died.
ssize_t got = ::recvmsg(m_relay[1], &message, MSG_DONTWAIT | MSG_CMSG_CLOEXEC);
if (got < 0 && (errno == EAGAIN || errno == EWOULDBLOCK)) {
struct pollfd waiting = {};
waiting.fd = m_relay[1];
waiting.events = POLLIN;
if (::poll(&waiting, 1, kRelayWaitMs) > 0) {
// The kernel rewrites these on every call, so a retry starts from the same state as
// the first attempt.
message.msg_controllen = sizeof(control);
message.msg_flags = 0;
got = ::recvmsg(m_relay[1], &message, MSG_DONTWAIT | MSG_CMSG_CLOEXEC);
}
}
if (got < 0) {
MGLOG_W_ONCE("gbm frame channel: a relayed handle could not be read back (%d: %s); the frame is not "
"offered to the container",
errno, std::strerror(errno));
DropRelay();
return false;
}
if (got == 0) {
// Zero bytes on a connected socket is the end of it. Only this process holds either end
// of this socketpair, so this is a bug being reported rather than a peer going away.
MGLOG_W_ONCE("gbm frame channel: the handle relay is closed; the frame is not offered to the container");
DropRelay();
return false;
}
// The descriptors come out of the control data before anything is decided about the
// message, because they belong to this process from the moment they were received: a path
// that returns without closing them leaks a descriptor per frame.
Int collected = 0;
for (struct cmsghdr* header = CMSG_FIRSTHDR(&message); header != nullptr;
header = CMSG_NXTHDR(&message, header)) {
if (header->cmsg_level != SOL_SOCKET || header->cmsg_type != SCM_RIGHTS) continue;
const SizeT bytes = (SizeT)header->cmsg_len - CMSG_LEN(0);
const Int count = (Int)(bytes / sizeof(Int));
const Int* descriptors = reinterpret_cast<const Int*>(CMSG_DATA(header));
for (Int i = 0; i < count; ++i) {
if (collected < kMaxHandleFds) {
fds[collected++] = descriptors[i];
} else {
// More descriptors than a frame can carry: they are this process's now, so they
// are closed here rather than leaked.
CloseQuietly(descriptors[i]);
}
}
}
if ((message.msg_flags & MSG_CTRUNC) != 0) {
MGLOG_W_ONCE("gbm frame channel: a frame's control data did not fit in %zu bytes; the frame is not "
"offered because its descriptor list is incomplete",
sizeof(control));
for (Int i = 0; i < collected; ++i) CloseQuietly(fds[i]);
DropRelay();
return false;
}
if ((message.msg_flags & MSG_TRUNC) != 0 || (SizeT)got < sizeof(GbmNativeHandleHeader)) {
// A message longer than a handle can be, or shorter than the three counts every handle
// begins with. Either way it cannot describe a frame, and a relay that has been read
// past the end of one message must not carry the next frame's header: it is recreated.
MGLOG_W_ONCE("gbm frame channel: a relayed handle is %zd bytes and %s; the frame is not offered", got,
((message.msg_flags & MSG_TRUNC) != 0) ? "longer than a handle can be"
: "too short to hold its own counts");
for (Int i = 0; i < collected; ++i) CloseQuietly(fds[i]);
DropRelay();
return false;
}
// THE BYTES THAT COME BACK ARE THE HANDLE'S INTEGER ARRAY, NOT A handle AND ITS COUNTS.
// Measured on the phone: parsing the front of this payload as {version, numFds, numInts}
// produced a NumFds taken from the buffer's own integer data, and every frame was refused
// with "its handle carries more descriptors than either end takes" - which is what the
// container saw as a connection that was accepted and then closed without a frame. The two
// counts are known HERE and are not in those bytes: the descriptors came out of the control
// data (collected of them) and the payload is their integer array. So the header is BUILT
// rather than copied, and what leaves this process is a real native_handle_t layout -
// version, numFds, numInts, then the ints - which is exactly what the container's parser
// reads (GbmInternal.h's AHBNativeHandleHeader).
if (payloadBytes + sizeof(GbmNativeHandleHeader) > capacity) {
MGLOG_W_ONCE("gbm frame channel: a handle's integer array does not fit in a frame message with its "
"header; the frame is not offered to the container");
for (Int i = 0; i < collected; ++i) CloseQuietly(fds[i]);
DropRelay();
return false;
}
std::memmove(payload + sizeof(GbmNativeHandleHeader), payload, payloadBytes);
GbmNativeHandleHeader header = {};
header.Version = (Int32)sizeof(GbmNativeHandleHeader);
header.NumFds = (Int32)collected;
header.NumInts = (Int32)(payloadBytes / sizeof(Int32));
std::memcpy(payload, &header, sizeof(header));
payloadBytes += sizeof(header);
fdCount = collected;
return true;
}
// The acceptor. It touches the listener - its own copy of the descriptor, which nothing else
// closes while this loop runs - and, once per dialer, the channel under its lock.
void GbmFrameChannel::AcceptLoop(Int listen) {
while (m_accepting.load(std::memory_order_acquire)) {
struct pollfd waiting = {};
waiting.fd = listen;
waiting.events = POLLIN;
const Int ready = ::poll(&waiting, 1, kAcceptPollMs);
if (ready < 0) {
if (errno == EINTR) continue;
MGLOG_W_ONCE("gbm frame channel: poll on the listening socket failed (%d: %s); no container can be "
"taken in until the socket is opened again",
errno, std::strerror(errno));
return;
}
if (ready == 0) continue;
// accept4, so the connection is close-on-exec and non-blocking from the moment it
// exists: a container that stops reading must not be able to park a send on either
// thread.
const Int fd = ::accept4(listen, nullptr, nullptr, SOCK_CLOEXEC | SOCK_NONBLOCK);
if (fd < 0) {
if (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK || errno == ECONNABORTED) continue;
MGLOG_W_ONCE("gbm frame channel: accept failed (%d: %s); no container can be taken in until the "
"socket is opened again",
errno, std::strerror(errno));
return;
}
const std::lock_guard<std::mutex> lock(m_mutex);
AdoptLocked(fd);
}
}
void GbmFrameChannel::AdoptLocked(Int fd) {
if (ClientAliveLocked()) {
// ONE CLIENT, which is what the protocol is written for, and a dialer that arrives while
// a live one is connected is closed rather than left connected - it learns it is not the
// one being served instead of holding a connection whose frames nobody reads.
CloseQuietly(fd);
MGLOG_W_ONCE("gbm frame channel: a second peer dialled %s while a container is connected; it is closed "
"and the frames stay with the connected one",
m_path.c_str());
return;
}
m_client = fd;
MGLOG_I("gbm frame channel: a container is connected on %s; every frame the host offers is sent to it as "
"the dma-buf descriptor behind it",
m_path.c_str());
// AND THE FRAME IT DIALED FOR GOES OUT FROM HERE. A compositor creates its GBM device before
// it asks for a surface, so the frame that is the draw target right now was offered before
// this connection existed; a container that is not told about it allocates every buffer of
// its new swapchain locally, and keeps them for the life of the process.
ReofferFrameLocked();
}
Bool GbmFrameChannel::ClientAliveLocked() {
if (m_client < 0) return false;
Uint8 peek = 0;
// MSG_PEEK, because a release waiting in the socket belongs to the render path: this test
// only asks whether the peer is still there, and a release it consumed would be a frame the
// channel believed was still outstanding.
const ssize_t got = ::recv(m_client, &peek, sizeof(peek), MSG_DONTWAIT | MSG_PEEK);
if (got > 0) return true;
if (got == 0) {
// Zero bytes on a connected socket is the end of it, whether the container exited or
// the kernel gave up on it. A compositor that crashed and is being restarted is
// exactly this, and its successor is the dialer standing right here.
MarkDown("the container closed the connection", 0);
return false;
}
if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR) return true;
MarkDown("recv", errno);
return false;
}
void GbmFrameChannel::MarkDown(const char* what, Int error) {
if (m_client < 0) return;
CloseQuietly(m_client);
m_client = -1;
if (error != 0) {
MGLOG_W("gbm frame channel: the container connection is down: %s failed (%d: %s); the frames are still "
"drawn and presented, and the container stops receiving them until it dials again",
what, error, std::strerror(error));
} else {
MGLOG_W("gbm frame channel: the container connection is down: %s; the frames are still drawn and "
"presented, and the container stops receiving them until it dials again",
what);
}
}
Bool GbmFrameChannel::OfferFrame(struct AHardwareBuffer* buffer, const HostFrameOffer& offer) {
if (buffer == nullptr) return false;
// Open() does nothing once the socket exists, so this is an integer test in the common case.
Open();
const std::lock_guard<std::mutex> lock(m_mutex);
// The releases the container has sent since the last frame, which is also how a container
// that died is noticed. Whoever replaced it was taken by the acceptor's own thread the
// moment it dialled: that thread is what makes a container's connect() succeed at all.
PollLocked();
// AND THE FRAME IS REMEMBERED BEFORE IT IS SENT. A container that dials while this call is
// running is one the acceptor hands the frame it finds here to, so the two threads must not
// be able to disagree about which frame is the one being drawn into.
const Uint32 seq = m_nextSeq;
RememberFrameLocked(buffer, offer, seq);
if (m_client < 0) return false;
return SendFrameLocked(buffer, offer, seq, /*counting=*/true);
}
void GbmFrameChannel::RememberFrameLocked(struct AHardwareBuffer* buffer, const HostFrameOffer& offer, Uint32 seq) {
if (m_frame == buffer) {
// The host redrawing into the buffer it already handed over: the same frame under a new
// sequence number. Acquiring it again would be a reference this channel takes and never
// gives back.
m_frameOffer = offer;
m_frameSeq = seq;
return;
}
// The reference is this channel's own, and it is what makes the frame a container that dials
// late is owed safe to hold: the backend drops its handle the moment the frame is presented,
// and a frame owed to a dialer arriving in the same millisecond may not be memory that has
// already gone back to the host's pool.
AHardwareBuffer_acquire(buffer);
ReleaseFrameLocked();
m_frame = buffer;
m_frameOffer = offer;
m_frameSeq = seq;
}
void GbmFrameChannel::ReleaseFrameLocked() {
if (m_frame == nullptr) return;
AHardwareBuffer_release(m_frame);
m_frame = nullptr;
m_frameOffer = {};
m_frameSeq = 0;
}
void GbmFrameChannel::DropFrame() {
const std::lock_guard<std::mutex> lock(m_mutex);
ReleaseFrameLocked();
}
void GbmFrameChannel::ReofferFrameLocked() {
if (m_frame == nullptr) return;
// The same frame, under the sequence number it was offered with: it IS the frame that is the
// draw target now, and a second number for it would be this channel claiming the host had
// handed over a frame it never handed over.
MGLOG_I("gbm frame channel: a container dialled after frame %u was offered; it is sent that frame, which is "
"still the draw target",
m_frameSeq);
SendFrameLocked(m_frame, m_frameOffer, m_frameSeq, /*counting=*/false);
}
Bool GbmFrameChannel::SendFrameLocked(struct AHardwareBuffer* buffer, const HostFrameOffer& offer, Uint32 seq,
Bool counting) {
if (m_client < 0) return false;
Uint8 payload[kMessageMax];
SizeT payloadBytes = 0;
Int fds[kMaxHandleFds];
for (Int i = 0; i < kMaxHandleFds; ++i) fds[i] = -1;
Int fdCount = 0;
if (!RelayHandle(buffer, payload, sizeof(payload), payloadBytes, fds, fdCount)) return false;
GbmNativeHandleHeader handle = {};
std::memcpy(&handle, payload, sizeof(handle));
const SizeT tail = ((SizeT)handle.NumFds + (SizeT)handle.NumInts) * sizeof(Int32);
const SizeT messageBytes = sizeof(GbmFrameOffer) + sizeof(GbmNativeHandleHeader) + tail;
// EVERY REASON THIS CHANNEL REFUSES A FRAME, IN ONE PLACE. Each of them is a frame the
// container would refuse anyway - it checks the same counts and the same geometry - and a
// frame that is refused here costs nothing: no message, no sequence number, and no entry in
// the count of frames waiting to be released.
const char* refusal = nullptr;
if (offer.Width == 0 || offer.Height == 0 || offer.StrideBytes == 0) {
refusal = "the host described it with a width, height or stride of zero, and a buffer with no size is "
"not one a compositor can render into";
} else if (handle.NumFds < 1) {
refusal = "its handle carries no descriptor, and a frame without a dma-buf is not one the container "
"can render into";
} else if (handle.NumFds > kMaxHandleFds) {
refusal = "its handle carries more descriptors than either end takes";
} else if (handle.NumInts < 0) {
refusal = "its handle claims a negative number of integers";
} else if ((SizeT)handle.NumInts > kMessageMax / sizeof(Int32)) {
// This is what keeps the message's length arithmetic from wrapping on a 32 bit ABI: a
// handle whose integer array alone is longer than a whole message cannot be carried by
// this protocol whatever the rest of it says, and the check below would otherwise be
// reading a number that had already turned into a small and plausible one.
refusal = "its handle names more integers than a whole message could carry";
} else if (fdCount < handle.NumFds) {
refusal = "fewer descriptors arrived with it than its handle names";
} else if (messageBytes > kMessageMax) {
refusal = "the message it makes is longer than the container reads";
}
Bool offered = false;
if (refusal != nullptr) {
// ONCE, not per frame: a host that describes every frame the same broken way must not
// turn this into a line per frame on the apply thread.
MGLOG_W_ONCE("gbm frame channel: frame %u (%ux%u) is not offered to the container: %s", offer.Index,
offer.Width, offer.Height, refusal);
} else {
// One message, built whole: the offer, then the handle's own three counts, then the two
// arrays the counts describe. What came back from the relay is copied in as far as it
// goes and the remainder is left zero, because the container checks the LENGTH it
// derives from those counts and never reads the arrays themselves - it takes the
// descriptors from the control data and hands the first one to GBM. So a handle whose
// bytes are shorter than its own counts imply is padded rather than refused, and one
// longer than they imply is cut to what the counts say; either way the length the
// container computes is the length it receives.
Uint8 message[kMessageMax] = {};
GbmFrameOffer out = {};
out.Magic = kGbmFrameOfferMagic;
out.Version = kGbmFrameVersion;
out.Seq = seq;
out.Index = offer.Index;
out.Width = offer.Width;
out.Height = offer.Height;
out.StrideBytes = offer.StrideBytes;
out.Format = offer.Format;
out.UsageLo = offer.UsageLo;
out.UsageHi = offer.UsageHi;
std::memcpy(message, &out, sizeof(out));
const SizeT handleBytes = (payloadBytes < sizeof(GbmNativeHandleHeader) + tail)
? payloadBytes
: (sizeof(GbmNativeHandleHeader) + tail);
std::memcpy(message + sizeof(out), payload, handleBytes);
struct iovec vector = {};
vector.iov_base = message;
vector.iov_len = messageBytes;
Uint8 control[CMSG_SPACE(sizeof(Int) * (SizeT)kMaxHandleFds)] = {};
struct msghdr outgoing = {};
outgoing.msg_iov = &vector;
outgoing.msg_iovlen = 1;
outgoing.msg_control = control;
outgoing.msg_controllen = CMSG_SPACE(sizeof(Int) * (SizeT)handle.NumFds);
struct cmsghdr* rights = CMSG_FIRSTHDR(&outgoing);
rights->cmsg_level = SOL_SOCKET;
rights->cmsg_type = SCM_RIGHTS;
rights->cmsg_len = CMSG_LEN(sizeof(Int) * (SizeT)handle.NumFds);
std::memcpy(CMSG_DATA(rights), fds, sizeof(Int) * (SizeT)handle.NumFds);
// MSG_DONTWAIT for the promise at the top of this file, and MSG_NOSIGNAL because a
// server killed by SIGPIPE when the container goes away takes the frames it was drawing
// with it. A full socket, a reset peer and a broken pipe all say the same thing here:
// this connection can no longer be trusted with a frame, so it is closed and the next
// container to dial is served.
const ssize_t sent = ::sendmsg(m_client, &outgoing, MSG_DONTWAIT | MSG_NOSIGNAL);
if (sent != (ssize_t)messageBytes) {
MarkDown("sendmsg", (sent < 0) ? errno : 0);
} else {
// A frame offered for the first time is the next sequence number and one more frame
// outstanding. The same frame sent again to a container that has just dialled is
// neither: the host handed over nothing new, and a count of frames waiting to be
// released may not be moved by a message that describes one already in it.
if (counting) {
m_nextSeq = seq + 1;
++m_offered;
++m_outstanding;
}
offered = true;
// The first offer is the one worth a line, and it carries the host's own numbers:
// the format in particular is the field the two ends could disagree about without
// either failing, so the value the host actually sends is put where a run on a
// device can be read afterwards.
MGLOG_I_ONCE("gbm frame channel: the first frame is offered to the container (seq %u index %u %ux%u "
"stride %u format 0x%08x)",
out.Seq, out.Index, out.Width, out.Height, out.StrideBytes, out.Format);
MGLOG_D("gbm frame channel: frame seq=%u index=%u offered (%zu bytes, %d descriptor(s), %llu "
"outstanding)",
out.Seq, out.Index, messageBytes, handle.NumFds, (unsigned long long)m_outstanding);
}
}
// The kernel took its own copies of these when the message was sent, and one that was not
// sent is not the container's to hold: every path closes what the relay handed over.
for (Int i = 0; i < fdCount; ++i) CloseQuietly(fds[i]);
return offered;
}
void GbmFrameChannel::Poll() {
const std::lock_guard<std::mutex> lock(m_mutex);
PollLocked();
}
void GbmFrameChannel::PollLocked() {
if (m_client < 0) return;
for (Int drained = 0; drained < kDrainMax; ++drained) {
GbmFrameRelease release = {};
struct iovec vector = {};
vector.iov_base = &release;
vector.iov_len = sizeof(release);
struct msghdr message = {};
message.msg_iov = &vector;
message.msg_iovlen = 1;
// recvmsg and not recv, for MSG_TRUNC: a message longer than a release can only be a
// peer speaking something else, and a release read out of the first twelve bytes of it
// would name a frame nobody released.
const ssize_t got = ::recvmsg(m_client, &message, MSG_DONTWAIT);
if (got < 0) {
if (errno == EINTR) continue;
// Nothing said, which is what a container that is between frames looks like.
if (errno == EAGAIN || errno == EWOULDBLOCK) return;
MarkDown("recvmsg", errno);
return;
}
if (got == 0) {
// Zero bytes on a connected socket is the end of it, whether the container exited or
// the kernel gave up on it. Its replacement is taken by the next frame.
MarkDown("the container closed the connection", 0);
return;
}
if ((message.msg_flags & MSG_TRUNC) != 0 || got != (ssize_t)sizeof(release)) {
// The message is skipped, not the connection: on a message-oriented socket a
// message too long for the buffer is consumed whole, so the next one is still a
// message rather than the tail of this one.
MGLOG_W_ONCE("gbm frame channel: the container sent a message that is not a %zu byte release; it is "
"skipped",
sizeof(release));
continue;
}
if (release.Magic != kGbmFrameReleaseMagic) {
MGLOG_W_ONCE("gbm frame channel: a message from the container begins with 0x%08x rather than 'MGGR'; "
"it is skipped",
release.Magic);
continue;
}
if (release.Version != kGbmFrameVersion) {
// Refused rather than acted on: the version is what says how the rest of the message
// is laid out, and a release read with the wrong layout names a frame the container
// did not mean.
MGLOG_E_ONCE("gbm frame channel: the container speaks protocol version %u where this server speaks "
"version %u; its releases are refused until the two agree",
release.Version, (Uint32)kGbmFrameVersion);
continue;
}
++m_released;
if (m_outstanding > 0) --m_outstanding;
// The first release is the only one worth a line: it is the proof that the container
// took the descriptor and its compositor retired the frame, where every later one says
// the same thing sixty times a second.
MGLOG_I_ONCE("gbm frame channel: the container is releasing frames (first: seq %u)", release.Seq);
}
// Anything past this cap stays in the socket, whose buffer the kernel bounds, and goes out
// with the next frame: a peer that says nothing but releases may not spend the apply thread
// here.
}
void GbmFrameChannel::Close() {
// THE ACCEPTOR IS STOPPED AND JOINED FIRST, and outside the lock. It may be inside that
// lock when this is called - a teardown that waited for it while holding it would wait for
// ever - and it polls the listening descriptor, which may not be closed under it: the next
// open() in this process hands out the descriptor number a moment later, and the thread
// would then be polling somebody else's socket.
m_accepting.store(false, std::memory_order_release);
if (m_acceptor.joinable()) m_acceptor.join();
const std::lock_guard<std::mutex> lock(m_mutex);
// The frame is this channel's reference and is released with it: the backend has already
// dropped its own by the time a teardown here is a teardown of the frame as well.
ReleaseFrameLocked();
// The connection is closed quietly rather than marked down: this is a teardown, not a
// failure, and a warning about it would be noise in a log read for the failures.
CloseQuietly(m_client);
m_client = -1;
DropRelay();
if (m_listen >= 0) {
// The node is this channel's to remove only while it still IS this channel's socket: a
// server that was restarted has bound the same path with a socket of its own, and a
// teardown that unlinked it would take the successor's node away with it. st_dev and
// st_ino against the listening descriptor is the cheapest way to ask, and a check that
// fails only leaves a name behind that the next Open() unlinks anyway.
struct stat node = {};
struct stat socket = {};
if (!m_path.empty() && ::stat(m_path.c_str(), &node) == 0 && ::fstat(m_listen, &socket) == 0 &&
node.st_dev == socket.st_dev && node.st_ino == socket.st_ino) {
::unlink(m_path.c_str());
}
CloseQuietly(m_listen);
m_listen = -1;
}
m_path.clear();
// The counters are history and stay: a caller that logs Outstanding() after a teardown is
// asking what happened, and IsUp() is what answers what is connected now.
}
}
#else // !(MOBILEGL_BUILD_DISAGGREGATED && __ANDROID__)
namespace MobileGL::MG_Backend {
// A build that is not the Android render server has no AHardwareBuffer to offer and no
// container beside it. The class exists so that callers do not have to be compiled
// differently, and every entry point answers the way an absent container does.
GbmFrameChannel::~GbmFrameChannel() = default;
Bool GbmFrameChannel::Open() { return false; }
Bool GbmFrameChannel::OfferFrame(struct AHardwareBuffer*, const HostFrameOffer&) { return false; }
void GbmFrameChannel::Poll() {}
void GbmFrameChannel::Close() {}
}
#endif
// End of Source File Header
+281
View File
@@ -0,0 +1,281 @@
// MobileGL - MobileGL/MG_Backend/GbmFrameChannel.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Includes.h>
#include "HostFrameBridge.h"
struct AHardwareBuffer;
namespace MobileGL::MG_Backend {
// THE FRAME THE HOST HANDED OVER, HANDED ON TO THE CONTAINER AS A dma-buf.
//
// The display host owns a pool of AHardwareBuffers: it allocates them, this side imports the
// one it offers as a draw target (HostFrameBridge, DirectGLES::HostFrameTarget), and the host
// puts it on the glass when this side answers that it is drawn. The Wayland compositor in the
// Linux container has to render into THOSE SAME frames - it runs this project's own GBM, and a
// gbm_bo_create there answers with the phone's frame instead of a dma-heap allocation of its
// own - so every frame this side takes has to reach the container as that frame's dma-buf
// descriptor. That, and the release that comes back, is all this channel carries.
//
// The container's end is MobileGL/MG_Gbm/GbmHostChannel.cpp. It reads exactly the bytes
// declared below and answers with a release once its compositor no longer reads the frame.
// The three shapes ARE that file's GbmFrameOffer/GbmFrameRelease and GbmInternal.h's
// AHBNativeHandleHeader, field for field and packing included. They are restated here rather
// than shared because the two ends are built into different libraries from different build
// configurations: MG_Backend is the Android render server and never sees libgbm's headers,
// while MG_Gbm is built for the container and never sees Android's. A change on either side
// is a protocol change, and the magic is what makes a mismatched pair fail loudly instead of
// being read as plausible numbers.
//
// ONE MESSAGE PER FRAME, on SOCK_SEQPACKET, and that is part of the protocol rather than an
// implementation detail: an offer and its handle only mean anything together, and the message
// boundary is what lets the reader take the two as one thing. The container derives the
// length of the message it expects from the handle's own two counts and refuses any other
// length, so this end sends the offer immediately followed by the handle's counts and its two
// arrays, with the descriptors in the SCM_RIGHTS control data of that same sendmsg.
//
// The channel is OPTIONAL, and that is the rule the implementation is built around: a
// container that is not running, has died, or has stopped reading may cost the render path a
// syscall or two, and may never block it, fail it, or grow without bound. Every entry point
// below is non-blocking and answers silence with silence.
inline constexpr Uint32 kGbmFrameOfferMagic = 0x4647474Du; // 'M','G','G','F' in memory order
inline constexpr Uint32 kGbmFrameReleaseMagic = 0x5247474Du; // 'M','G','G','R' in memory order
inline constexpr Uint32 kGbmFrameVersion = 1u;
// WHERE THE CONTAINER DIALS. /data/local/tmp is the one directory both halves of a split run
// can be pointed at on a device, and the default lives on this end because this end is the
// server. The container's compiled-in default is /run/mobilegl-gbm.sock - the same protocol
// with the path each side can reach unaided - so a deployment that puts the two in different
// mount namespaces sets MOBILEGL_GBM_SOCKET on both; it is read here and in
// GbmHostChannelCreate, from the same variable, for that reason.
inline constexpr const char* kGbmFrameSocketPath = "/data/local/tmp/mobilegl-gbm.sock";
inline constexpr const char* kGbmFrameSocketEnv = "MOBILEGL_GBM_SOCKET";
// One frame the host offered, described for the container. Width, Height, StrideBytes and
// Format are the host's own numbers for the frame (HostFrameOffer), carried through
// unchanged: they describe the memory the descriptor points at, and this end is not the place
// where a number the other two ends agreed on gets rewritten.
struct GbmFrameOffer {
Uint32 Magic;
Uint32 Version;
Uint32 Seq; // this channel's, one per frame offered
Uint32 Index; // the host's own buffer index, echoed so the phone can match a frame
Uint32 Width;
Uint32 Height;
Uint32 StrideBytes;
Uint32 Format;
Uint32 UsageLo;
Uint32 UsageHi;
} __attribute__((packed));
// What the container says once it is done with a frame. It may arrive late, it may never
// arrive, and both are cheap here: the channel keeps a count and no per-frame record (see
// Outstanding below), so nothing is leaked and nothing grows when one is missed.
struct GbmFrameRelease {
Uint32 Magic;
Uint32 Version;
Uint32 Seq;
} __attribute__((packed));
// THE OTHER DIRECTION: A dma-buf THE CONTAINER WANTS THIS SIDE TO BACK A TEXTURE WITH.
//
// The compositor's scene reaches the display daemon's dma-bufs through glEGLImageTargetTexture2DOES
// (MG_Impl/GLImpl/Exporting/Definitions.cpp, MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp both
// name this half as the next change). eglCreateImageKHR has already held a dup() of every plane's
// descriptor since the image was made; what was missing is the render server's half - telling THIS
// side which texture that dma-buf backs, over the one channel that can carry a descriptor across the
// container boundary. This message is that half: the client names the texture target and the
// texture's own name, describes the buffer exactly as drmModeAddFB2 would, and the plane descriptors
// ride in the SCM_RIGHTS control data of the same sendmsg. One message per import, as SEQPACKET
// requires, so a descriptor and its description are never read apart.
//
// The reply is what the client waits for before a framebuffer is allowed to be complete over the
// texture: Ok=0 means this side could not make the import real, and the client keeps treating the
// texture as having no storage rather than letting it render into nothing.
inline constexpr Uint32 kGbmImportRequestMagic = 0x4947474Du; // 'M','G','G','I' in memory order
inline constexpr Uint32 kGbmImportReplyMagic = 0x52474749u; // 'I','G','G','R' in memory order
inline constexpr Uint32 kGbmImportVersion = 1u;
inline constexpr Uint32 kGbmImportMaxPlanes = 4u;
struct GbmImportRequest {
Uint32 Magic;
Uint32 Version;
Uint32 Target; // the GL texture target the image was bound to (0x8D65 = GL_TEXTURE_EXTERNAL_OES)
Uint32 TexName; // the CLIENT's texture name; this side resolves its twin
Uint32 Width;
Uint32 Height;
Uint32 FourCC;
Uint32 PlaneCount; // 1..kGbmImportMaxPlanes; the descriptors are the message's control data
Int32 Stride[kGbmImportMaxPlanes];
Int32 Offset[kGbmImportMaxPlanes];
Uint64 Modifier[kGbmImportMaxPlanes];
Uint32 HasModifier[kGbmImportMaxPlanes];
} __attribute__((packed));
struct GbmImportReply {
Uint32 Magic;
Uint32 Version;
Uint32 Ok;
Uint32 Reserved;
} __attribute__((packed));
// An AHardwareBuffer native handle as it appears on the wire: the platform's native_handle_t
// (version, then the two counts, then the descriptor array followed by the integer array),
// written by AHardwareBuffer_sendHandleToUnixSocket and read by the container's
// AHBNativeHandleHeader. The counts are 32 bit on every Android ABI, which is why they are
// spelled out here rather than taken from a platform header this library cannot include.
struct GbmNativeHandleHeader {
Int32 Version;
Int32 NumFds;
Int32 NumInts;
} __attribute__((packed));
// ONE CONTAINER, ONE LISTENING SOCKET, TWO THREADS: the render path's, and an acceptor's.
//
// A compositor dials ONCE - when it creates its GBM device - and keeps that connection for the
// life of the process, while this side offers a frame per drawn frame. An accept that only ran
// inside OfferFrame therefore ran after the dial it was supposed to answer, and a dialer that
// nobody took kept the listener's queue full for every later one. That is measured on the
// phone rather than argued: with accept() only on the offer path, connect() from the container
// answered EAGAIN(11) on four attempts out of four while the node sat there listening.
//
// So the accept has a thread of its own, started by Open() and stopped by Close(), and it
// adopts what it takes on the spot - including sending it the frame the host last handed over,
// because a compositor's first allocation is the one its swapchain is built from. The
// connected socket, the frame that is remembered, the relay and the counters are shared between
// the two threads and guarded by the one mutex below; both entry points are non-blocking, so
// neither thread can park the other for longer than a syscall.
class GbmFrameChannel {
public:
GbmFrameChannel() = default;
GbmFrameChannel(const GbmFrameChannel&) = delete;
GbmFrameChannel& operator=(const GbmFrameChannel&) = delete;
~GbmFrameChannel();
// Creates the listening socket if it is not there yet and starts the acceptor, and does
// nothing otherwise. It never waits for a container: the socket is bound, a backlog is
// given to listen(), and the acceptor takes whoever dials from then on - whether or not a
// frame is being drawn. A caller that wants the container to find the socket before the
// first frame - the container dials once at its own device creation and retries about once
// a second - can call this when the host-framed surface is initialized, beside
// HostFrameBridge::Open.
Bool Open();
// Offers one frame the host just handed over, as the dma-buf descriptor behind it.
// False means the frame did not go out, for any of the reasons this channel treats as
// ordinary: no container is connected, the container just died, or the frame could not be
// described. None of them is the caller's problem and none of them is fatal - the frame
// is drawn and presented exactly as it would have been without a channel at all.
Bool OfferFrame(struct AHardwareBuffer* buffer, const HostFrameOffer& offer);
// Reads whatever the container has said since the last call and marks the channel down if
// the connection has ended. OfferFrame already does this once per frame, so a caller only
// needs it if it wants releases drained on a path that has no new frame in it.
void Poll();
// The frame the host handed over is gone - it was presented, or it was replaced - and this
// channel may no longer hold it or offer it to anybody. THE CHANNEL ITSELF STAYS OPEN, and
// that is the point of the call: the listening socket, its node and the container's
// connection outlive every frame, and a channel that was closed and reopened per frame took
// the node away sixty times a second - a container dialling at the wrong moment got ENOENT
// for a socket that was there a millisecond earlier and would be there a millisecond later.
void DropFrame();
// True while a container is connected and the frames are being offered to it.
Bool IsUp() const {
const std::lock_guard<std::mutex> lock(m_mutex);
return m_client >= 0;
}
// How many frames have been offered and not released, as a count of messages and not a
// record of them. It is here to be logged, and it is deliberately the only per-frame state
// the channel keeps: a release that never arrives costs a number that is too high, and
// nothing else, which is what makes a container that stops reading harmless.
Uint64 Outstanding() const {
const std::lock_guard<std::mutex> lock(m_mutex);
return m_outstanding;
}
// Closes the connection, the listening socket and the node it created. Idempotent, safe
// from a destructor, and safe to call while a container is connected: the container sees
// the connection end and goes back to allocating its frames locally, which is the state it
// starts in.
void Close();
private:
// The socketpair a frame's handle is handed to and read back from. Recreated on demand,
// because the one call the NDK offers for a buffer's descriptors sends them as a message
// of its own and the container takes one message per frame - see the note in the .cpp.
Bool EnsureRelay();
void DropRelay();
Bool RelayHandle(struct AHardwareBuffer* buffer, Uint8* payload, SizeT capacity, SizeT& payloadBytes,
Int* fds, Int& fdCount);
// Takes the one container the protocol allows, and closes anyone else who dials while a
// LIVE one is connected rather than leaving them to believe they are being served. A
// connection whose peer has exited is not a live one: it is closed here and its dialer's
// replacement is taken, which is what a compositor that crashed and was restarted is.
void AdoptLocked(Int fd);
// True while the connected container is still there. recvmsg with MSG_PEEK is what asks:
// zero bytes is the peer's close, EAGAIN is a peer that is between frames, and a release
// waiting in the socket is a peer that is alive - peeked at rather than read, because the
// render path is the one that spends releases.
Bool ClientAliveLocked();
// Takes the one container the protocol allows out of whoever dialled.
void AcceptLoop(Int listen);
// Sends the frame that is the draw target to a container that has just connected.
void ReofferFrameLocked();
// Remembers the frame the host last handed over, holding a reference to it so that a
// container which dials after the offer can still be given it. The previous one is
// released here; nothing else in the channel holds a frame.
void RememberFrameLocked(struct AHardwareBuffer* buffer, const HostFrameOffer& offer, Uint32 seq);
void ReleaseFrameLocked();
// The offer itself: the handle relayed out of the buffer, the message built, one sendmsg.
// A frame that has already been offered once is sent again with the sequence number it was
// offered under and does not move the counters, because it is the same frame and not a new
// one.
Bool SendFrameLocked(struct AHardwareBuffer* buffer, const HostFrameOffer& offer, Uint32 seq, Bool counting);
// The body of Poll(), for a caller that is already holding the lock.
void PollLocked();
// Closes the connection and says why, once. The channel stays open for the next container.
void MarkDown(const char* what, Int error);
Int m_listen = -1;
Int m_client = -1;
Int m_relay[2] = {-1, -1};
String m_path;
// Bumped only for a frame that actually went out, so a sequence number is never reused and
// a release can never be read as belonging to a frame of an earlier connection.
Uint32 m_nextSeq = 0;
Uint64 m_offered = 0;
Uint64 m_released = 0;
Uint64 m_outstanding = 0;
// The frame the host handed over most recently, held by this channel's own reference for as
// long as it is the draw target: it is what a container that dials late is sent.
struct AHardwareBuffer* m_frame = nullptr;
HostFrameOffer m_frameOffer = {};
Uint32 m_frameSeq = 0;
// The acceptor, and the flag that ends it. The listening descriptor is handed to the thread
// as its own copy, so it is the only thing in here that thread reads without the lock - and
// Close() joins the thread before it closes that descriptor.
std::thread m_acceptor;
std::atomic<Bool> m_accepting{false};
// Guards every member above, from both threads. Mutable because the two const questions a
// caller asks about the channel - is a container connected, how many frames are outstanding
// - are answers that have to be read under it.
mutable std::mutex m_mutex;
};
}
+189
View File
@@ -0,0 +1,189 @@
// MobileGL - MobileGL/MG_Backend/HostFrameBridge.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "HostFrameBridge.h"
#if MOBILEGL_BUILD_DISAGGREGATED && defined(__ANDROID__)
#include <android/hardware_buffer.h>
#include <cerrno>
#include <cstring>
#include <poll.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <unistd.h>
namespace MobileGL::MG_Backend {
namespace {
Bool ReadExactly(Int fd, void* out, SizeT n) {
Uint8* at = static_cast<Uint8*>(out);
while (n > 0) {
const ssize_t got = ::read(fd, at, n);
if (got <= 0) {
if (got < 0 && errno == EINTR) continue;
return false;
}
at += got;
n -= static_cast<SizeT>(got);
}
return true;
}
Bool WriteExactly(Int fd, const void* in, SizeT n) {
const Uint8* at = static_cast<const Uint8*>(in);
while (n > 0) {
const ssize_t put = ::write(fd, at, n);
if (put <= 0) {
if (put < 0 && errno == EINTR) continue;
return false;
}
at += put;
n -= static_cast<SizeT>(put);
}
return true;
}
}
HostFrameBridge::~HostFrameBridge() { Close(); }
Bool HostFrameBridge::Open(const char* path, Int acceptTimeoutMs, String& why) {
Close();
const char* name = (path != nullptr && path[0] != 0) ? path : kHostFrameSocket;
m_listen = ::socket(AF_UNIX, SOCK_STREAM, 0);
if (m_listen < 0) {
why = std::format("socket(AF_UNIX): {}", std::strerror(errno));
return false;
}
sockaddr_un address{};
address.sun_family = AF_UNIX;
if (std::strlen(name) >= sizeof(address.sun_path)) {
why = std::format("the socket name is longer than the kernel takes: {}", name);
Close();
return false;
}
// "@name" is the abstract namespace: sun_path[0] is NUL and the name follows it.
// Nothing is created on disk, so nothing is left behind and no file permission stands
// between two apps that may not write each other's directories.
if (name[0] == '@') {
address.sun_path[0] = 0;
std::strncpy(address.sun_path + 1, name + 1, sizeof(address.sun_path) - 2);
} else {
std::strncpy(address.sun_path, name, sizeof(address.sun_path) - 1);
// A leftover socket file from a server that died is not a reason to refuse the host.
::unlink(name);
}
if (::bind(m_listen, reinterpret_cast<sockaddr*>(&address), sizeof(address)) != 0) {
why = std::format("bind({}): {}", name, std::strerror(errno));
Close();
return false;
}
if (::listen(m_listen, 1) != 0) {
why = std::format("listen({}): {}", name, std::strerror(errno));
Close();
return false;
}
pollfd waiting{};
waiting.fd = m_listen;
waiting.events = POLLIN;
const Int ready = ::poll(&waiting, 1, acceptTimeoutMs);
if (ready <= 0) {
why = (ready == 0) ? std::format("no display host dialled {} within {} ms", name, acceptTimeoutMs)
: std::format("poll({}): {}", name, std::strerror(errno));
Close();
return false;
}
m_fd = ::accept(m_listen, nullptr, nullptr);
if (m_fd < 0) {
why = std::format("accept({}): {}", name, std::strerror(errno));
Close();
return false;
}
MGLOG_I("host frame bridge: a display host is on %s", name);
return true;
}
Bool HostFrameBridge::Acquire(struct AHardwareBuffer** buffer, HostFrameOffer& offer, String& why) {
*buffer = nullptr;
offer = HostFrameOffer{};
if (m_fd < 0) {
why = "the bridge is not open";
return false;
}
AHardwareBuffer* received = nullptr;
if (AHardwareBuffer_recvHandleFromUnixSocket(m_fd, &received) != 0 || received == nullptr) {
why = "the host stopped offering frames, or the handle did not cross";
return false;
}
if (!ReadExactly(m_fd, &offer, sizeof(offer)) || offer.Magic != kHostFrameMagic) {
// The handle is already this side's by the time this fails, so it goes back before
// the failure does: a reference held past the connection starves the host's pool.
AHardwareBuffer_release(received);
why = std::format("the frame's description did not follow its handle (magic 0x{:08x})", offer.Magic);
return false;
}
*buffer = received;
return true;
}
Bool HostFrameBridge::Complete(const HostFrameOffer& offer, Bool imported, Bool drawn, Bool fenceOk, String& why) {
if (m_fd < 0) {
why = "the bridge is not open";
return false;
}
HostFrameAck ack{};
ack.Magic = kHostFrameMagic;
ack.Version = kHostFrameVersion;
ack.Index = offer.Index;
ack.Imported = imported ? 1u : 0u;
ack.Drawn = drawn ? 1u : 0u;
ack.FenceOk = fenceOk ? 1u : 0u;
if (!WriteExactly(m_fd, &ack, sizeof(ack))) {
why = std::format("the answer to frame {} did not go out: {}", offer.Index, std::strerror(errno));
return false;
}
// The host answers with its own view of the pixels. It is what closes the round trip, so
// it is read here and kept rather than left in the socket for the next frame to trip over.
m_lastSeen = HostFrameSeen{};
if (!ReadExactly(m_fd, &m_lastSeen, sizeof(m_lastSeen)) || m_lastSeen.Magic != kHostFrameMagic) {
why = std::format("frame {} was answered but the host's readback did not arrive", offer.Index);
m_lastSeen = HostFrameSeen{};
return false;
}
return true;
}
void HostFrameBridge::Close() {
if (m_fd >= 0) ::close(m_fd);
if (m_listen >= 0) ::close(m_listen);
m_fd = -1;
m_listen = -1;
}
}
#else // !(MOBILEGL_BUILD_DISAGGREGATED && __ANDROID__)
namespace MobileGL::MG_Backend {
HostFrameBridge::~HostFrameBridge() = default;
Bool HostFrameBridge::Open(const char*, Int, String& why) {
why = "host frames are a disaggregated Android path";
return false;
}
Bool HostFrameBridge::Acquire(struct AHardwareBuffer**, HostFrameOffer&, String& why) {
why = "host frames are a disaggregated Android path";
return false;
}
Bool HostFrameBridge::Complete(const HostFrameOffer&, Bool, Bool, Bool, String& why) {
why = "host frames are a disaggregated Android path";
return false;
}
void HostFrameBridge::Close() {}
}
#endif
+103
View File
@@ -0,0 +1,103 @@
// MobileGL - MobileGL/MG_Backend/HostFrameBridge.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Includes.h>
struct AHardwareBuffer;
namespace MobileGL::MG_Backend {
// THE DISPLAY HOST'S FRAMES, AS THEY ARRIVE.
//
// The Android side that owns the screen allocates its frames itself and offers each one over
// an AF_UNIX socket: the handle travels ahead of a fixed-size description on the same
// connection, and the host then waits for this side to say the frame is drawn before it puts
// it on the glass. Why the host allocates rather than this side asking for the display's own
// buffers is measured, not chosen: AHardwareBuffer_createFromHandle accepts an
// AHardwareBuffer this process allocated and refuses one the display queue handed over
// (gralloc BAD_VALUE), so the only frames a render server can draw into are the ones the host
// made for it.
//
// The three message shapes below ARE the host's (its ahb_bridge.h), field for field and
// packing included. They are restated here rather than shared because the two ends are built
// by different projects; a change on either side is a protocol change, and the magic is what
// makes a mismatched pair fail loudly instead of being read as plausible numbers.
inline constexpr Uint32 kHostFrameMagic = 0x4D474C41u; // the host spells it "ALGM" in memory
inline constexpr Uint32 kHostFrameVersion = 1u;
// AN ABSTRACT NAME, not a path: the two ends are separate apps, and neither may create a
// socket file in /data/local/tmp (its directory grants others traverse only). An abstract
// name lives in the kernel namespace instead of the filesystem, so there is no file
// permission to get wrong - the same reason this project names its own endpoints @mgl-*.
inline constexpr const char* kHostFrameSocket = "@mobilegl-host-frame";
struct HostFrameOffer {
Uint32 Magic;
Uint32 Version;
Uint32 Index;
Uint32 Width;
Uint32 Height;
Uint32 StrideBytes;
Uint32 Format;
Uint32 UsageLo;
Uint32 UsageHi;
char Note[64];
} __attribute__((packed));
struct HostFrameAck {
Uint32 Magic;
Uint32 Version;
Uint32 Index;
Uint32 Imported;
Uint32 Drawn;
Uint32 FenceOk;
Uint8 Color[4];
char Note[64];
} __attribute__((packed));
struct HostFrameSeen {
Uint32 Magic;
Uint32 Version;
Uint32 Index;
Uint32 Locked;
Uint8 Observed[4];
Uint8 ObservedCorner[4];
char Note[64];
} __attribute__((packed));
// One host connection at a time, which is what the host offers: it dials, hands over a frame,
// waits for the answer, and hands over the next.
class HostFrameBridge {
public:
HostFrameBridge() = default;
HostFrameBridge(const HostFrameBridge&) = delete;
HostFrameBridge& operator=(const HostFrameBridge&) = delete;
~HostFrameBridge();
// Listens on `path` (the well-known name by default) and waits for the host. A host that
// never dials is answered with false and a reason, never an abort.
Bool Open(const char* path, Int acceptTimeoutMs, String& why);
// Takes the next frame. Ownership of the returned AHardwareBuffer is the caller's, who
// must hand it back through ReleaseOwned() once the frame is off the glass.
Bool Acquire(struct AHardwareBuffer** buffer, HostFrameOffer& offer, String& why);
// Tells the host what became of the frame it offered, and reads back what the host's own
// view of the pixels holds. `drawn` is what the host waits for before it presents.
Bool Complete(const HostFrameOffer& offer, Bool imported, Bool drawn, Bool fenceOk, String& why);
// The readback the host returned for the last completed frame, as the host reported it.
const HostFrameSeen& LastSeen() const { return m_lastSeen; }
void Close();
private:
Int m_fd = -1;
Int m_listen = -1;
HostFrameSeen m_lastSeen{};
};
}
+123 -53
View File
@@ -14,7 +14,9 @@
#if MOBILEGL_BUILD_DISAGGREGATED
#include <MG_Pipe/MGPipe.h>
#include <MG_Pipe/PipeApply.h>
#include <MG_Pipe/PipeMutation.h> // MGPipeForgetAllPublications, at the end of ShutdownSplitRoles
#include <MG_Remote/Client/ClientSession.h>
#include <MG_Remote/FatalFunnel.h> // InstallPipeSessionFailHook, installed at step 0 below
#include <MG_Remote/Server/ServerLoop.h>
#include <MG_Remote/Server/ServerSession.h>
#endif
@@ -80,9 +82,10 @@ namespace MobileGL::MG_Backend {
// DirectGLES consumes all thirteen migrated families (P2's 0..6, P3a's 7..8, P4a's
// 9..12): it registers the resource op table in Initialize()
// (BackendObject_DirectGLES.cpp:849) and reads every other family out of gPipeInputs.
// DirectVulkan registers NO resource ops - MGPipeSetResourceOps has exactly one caller
// in the whole tree and it is Managers.cpp:2594 - so bit 7 is CLEAR for it, which is
// the same fact ObjectSubsystemControlScenario already pins from the client side.
// DirectVulkan registers its wire resource op table off monolith only
// (VkBufferManager::RegisterWireResourceOps, VkBufferManager.cpp): under a transport it
// owns buffer stores and claims bit 7 (the P4a mask, see ConsumedSubsystemsFor below);
// under monolith it registers none and its buffers keep the legacy BufferBackendOps path.
// P5e (MG_Remote/CONTRACT-P5E.md §6.2). THE ONE CONSTANT THE WHOLE PHASE HANGS ON.
//
// kCapRunAheadApply says "this server applies an unbarriered record without reading
@@ -159,6 +162,14 @@ namespace MobileGL::MG_Backend {
Bool InitServerRoleCommon() {
using namespace MobileGL::MG_Remote;
// 0. the Magma wire funnels' route to Session::Fail (P7 wave 0). FIRST, before
// CreateBackend can reach a wire arm: a hook installed after the first death it
// was meant to catch is a hook that does nothing on the only run that mattered.
// Here rather than in ServerMain because BOTH server roles pass through this
// function - the inproc role and the spawn/TCP session child - and a per-transport
// install is how one of the two ends up without it.
InstallPipeSessionFailHook();
// 1. the SERVER role's private backend object, on the app thread, with no GL and no
// EGL. The context is created and made current later, on mgl-srv-apply, when the
// client's first eglMakeCurrent crosses as a blocking control request.
@@ -176,9 +187,9 @@ namespace MobileGL::MG_Backend {
// ZERO IS THE EXPLICIT ANSWER FOR P5, not an omission (ServerSession.h's block):
// every optional capability bit belongs to the package that owns its question, and
// withholding one leaves the legacy path running, which is the safe direction.
// kCapNeedsHostIndexBytes and kCapNeedsHostUboBytes must be 0 for the whole of P5
// by ruling - they are the only two things that ask for an MGHostSpan, and 0 is
// what keeps every one of them out of the first IPC frame (contract table 0).
// kCapNeedsHostIndexBytes, kCapNeedsHostUboBytes and kCapViewportArray are RESERVED
// (P8-F) and never set here: the first two are the only things that ask for an
// MGHostSpan, and 0 keeps every one of them off the wire (contract table 0).
//
// P5b t2 (CONTRACT-P5B.md §6.5) PUBLISHES THE ONE BIT P5b ADDS, and this is the
// only place that can: the question kCapBackendOwnsXfbCapture answers is "does the
@@ -211,6 +222,32 @@ namespace MobileGL::MG_Backend {
capBits |= MG_Pipe::kCapTimerQuery;
}
}
// P7 wave 2 package C, OQ-10 (CONTRACT-P7 §5.4): kCapResidentSubData, PUBLISHED
// OFF THE SERVER'S OWN WIRE RESOURCE TABLE AND NOTHING ELSE.
//
// The question the bit answers is the one BufferObject::LandBytesIntoResidentStore
// asks before it chooses between emitting `buffer_subdata_resident` (opcode 49) and
// falling back to the ordered in-place host memcpy: "does the side that will APPLY
// this record implement the resident arm at all?" Under a transport the client has
// no op table to probe - MGPipeResourceOpsHaveSubDataResident says so in as many
// words (MG_Impl/Pipe/PipeFill.cpp) and reads this bit instead - and until now
// nothing ever set it, so the answer was permanently "no" and BOTH backends fell
// back to the memcpy while both of them in fact register a SubDataResident arm
// (DirectGLES Managers.cpp's g_glesResourceOps, Magma VkBufferManager.cpp's
// g_vulkanWireResourceOps). Magma's implementation was dead code on the wire.
//
// BY THE TABLE, NEVER BY THE BACKEND ENUM, and that is ID-39's lesson rather than a
// style preference: a mask is a statement about THIS backend's registered table, so
// a bit derived from `ActiveBackendType == ...` would keep claiming the capability
// for a build, a flavour or a future backend whose table does not carry the arm -
// which is exactly how 66 uploads were lost to a consumer bit that nobody could
// have honoured. `g_resourceOpsAtStep2` is the pointer this function just pinned,
// and step 5 already refuses a run where that pointer was swapped underneath us, so
// reading the arm off it means the published bit and the table that has to answer
// for it cannot come apart.
if (g_resourceOpsAtStep2 != nullptr && g_resourceOpsAtStep2->SubDataResident != nullptr) {
capBits |= MG_Pipe::kCapResidentSubData;
}
// Each backend has an independent implementation-readiness gate.
// The runtime RunAhead knob can decline the feature, never create it.
const Bool runAheadReady = MG_Config::ActiveBackendType == BackendType::DirectVulkan
@@ -267,55 +304,16 @@ namespace MobileGL::MG_Backend {
pActiveBackendObject = MakeUnique<MG_Remote::Client::BackendObject_Remote>();
return true;
}
} // namespace
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
void ShutdownSplitRoles() {
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
// ClientSession::Stop IS table 3's whole order and it is idempotent: publish and wait
// for the server to drain (bounded - a lost record must be a red lane, not a hung
// exit), Doorbell::Kill through the transport's Shutdown, ServerLoop::Stop's bounded
// join - which also destroys the server's private BackendObject ON the apply thread
// while it still owns the context - the transport, and only THEN anything an emitter
// owns. A var-tail still named by an unapplied record is a use-after-free the join is
// what prevents, which is why the order is not a preference.
MG_Remote::Client::ClientSessionInstance().Stop();
// M-6: ClientSession::Stop's !m_started arm (a Start that FAILED after
// ServerSession::Accept - a refused Accept, an invalid cmd/reply ring) tears down only
// the client half and never stops the apply thread or drops the server's private backend,
// which ServerLoop::CreateBackend already built and which holds the process-wide
// g_resourceOps. So call ServerLoop::Stop() here unconditionally. It is idempotent: on the
// started path ClientSession::Stop already joined the thread, so this hits Stop's
// !joinable arm, which resets a backend that never ran a thread and is otherwise a no-op.
// Without this an early Start failure leaves BackendObject_DirectGLES permanently alive
// and every later split bring-up in the process fails at CreateBackend's m_backend!=null
// guard.
MG_Remote::Server::ServerLoopInstance().Stop();
}
#endif
// F1 (P7 wave 2). THE WHOLE SPLIT ARM OF Init(), AS ONE BODY, so it can run from
// either of the two places that may reach it - MobileGL::Initialize's early hook,
// before MG_State::Init(), and Init() itself for a direct caller such as a unit
// fixture - without a second copy. Two copies of a bring-up order is how the two
// orders come to differ on the day one of them is wrong.
Bool g_splitBackendBroughtUp = false;
#if MOBILEGL_BUILD_DISAGGREGATED
// The spawn server's entry into the SAME bring-up the inproc server role
// runs (ServerRole.h). A thin forwarder on purpose: the body stays in the
// anonymous namespace beside InitSplitRoles so the two cannot drift.
//
// THE WHOLE FUNCTION IS INSIDE THE GUARD, not just its body. A version with
// the guard inside still DEFINES the symbol in a pull build, and G1 caught
// it: 2 symbols added, 1 removed, .text +16 bytes. The pull build's identity
// is byte-for-byte, and "it returns false there" is not the same as "it is
// not there".
Bool InitServerRoleForSpawn() { return InitServerRoleCommon(); }
#endif
void Init() {
MGLOG_D("Initializing MobileGL Backend...");
#if MOBILEGL_BUILD_DISAGGREGATED
// THE SINGLE HOOK. In a build without MOBILEGL_BUILD_DISAGGREGATED, MG_Config::Transport
// is a `constexpr Monolith` (Config.h) and this whole statement is discarded, so the
// pull build gains no symbol, no branch and no byte - which is what G1 measures.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
void InitSplitBackend() {
g_splitBackendBroughtUp = true;
if (!InitSplitRoles()) {
// NOT a fallback to the switch. pActiveBackendObject stays null and the next GL
// call fails loudly, which is the only honest outcome: the operator asked for a
@@ -364,6 +362,78 @@ namespace MobileGL::MG_Backend {
}
}
LogBackendInfo();
}
} // namespace
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// F1 (P7 wave 2). See BackendObjects.h for why this exists and MobileGL/Init.cpp for the
// defect the order closes.
Bool InitSplitRolesBeforeState() {
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return false;
InitSplitBackend();
return true;
}
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
void ShutdownSplitRoles() {
g_splitBackendBroughtUp = false;
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
// ClientSession::Stop IS table 3's whole order and it is idempotent: publish and wait
// for the server to drain (bounded - a lost record must be a red lane, not a hung
// exit), Doorbell::Kill through the transport's Shutdown, ServerLoop::Stop's bounded
// join - which also destroys the server's private BackendObject ON the apply thread
// while it still owns the context - the transport, and only THEN anything an emitter
// owns. A var-tail still named by an unapplied record is a use-after-free the join is
// what prevents, which is why the order is not a preference.
MG_Remote::Client::ClientSessionInstance().Stop();
// M-6: ClientSession::Stop's !m_started arm (a Start that FAILED after
// ServerSession::Accept - a refused Accept, an invalid cmd/reply ring) tears down only
// the client half and never stops the apply thread or drops the server's private backend,
// which ServerLoop::CreateBackend already built and which holds the process-wide
// g_resourceOps. So call ServerLoop::Stop() here unconditionally. It is idempotent: on the
// started path ClientSession::Stop already joined the thread, so this hits Stop's
// !joinable arm, which resets a backend that never ran a thread and is otherwise a no-op.
// Without this an early Start failure leaves BackendObject_DirectGLES permanently alive
// and every later split bring-up in the process fails at CreateBackend's m_backend!=null
// guard.
MG_Remote::Server::ServerLoopInstance().Stop();
// The server's twins died with the session; the frontend objects that are destroyed after
// this (the GL context's) have nothing left on the other side to delete.
MG_Pipe::MGPipeForgetAllPublications();
}
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// The spawn server's entry into the SAME bring-up the inproc server role
// runs (ServerRole.h). A thin forwarder on purpose: the body stays in the
// anonymous namespace beside InitSplitRoles so the two cannot drift.
//
// THE WHOLE FUNCTION IS INSIDE THE GUARD, not just its body. A version with
// the guard inside still DEFINES the symbol in a pull build, and G1 caught
// it: 2 symbols added, 1 removed, .text +16 bytes. The pull build's identity
// is byte-for-byte, and "it returns false there" is not the same as "it is
// not there".
Bool InitServerRoleForSpawn() { return InitServerRoleCommon(); }
#endif
void Init() {
MGLOG_D("Initializing MobileGL Backend...");
#if MOBILEGL_BUILD_DISAGGREGATED
// THE SINGLE HOOK. In a build without MOBILEGL_BUILD_DISAGGREGATED, MG_Config::Transport
// is a `constexpr Monolith` (Config.h) and this whole statement is discarded, so the
// pull build gains no symbol, no branch and no byte - which is what G1 measures.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
// F1: MobileGL::Initialize now brings the split roles up BEFORE MG_State::Init(),
// through InitSplitRolesBeforeState() below, so by the time Init() is reached on
// that path the work is done. A DIRECT caller - a unit fixture that calls
// MG_Backend::Init() itself - still gets the whole bring-up here. The flag rather
// than `pActiveBackendObject == nullptr`, because a bring-up that FAILED leaves
// that pointer null too and retrying it would refuse at CreateBackend's
// m_backend != nullptr guard with a second, misleading line.
if (!g_splitBackendBroughtUp) InitSplitBackend();
return;
}
#endif
+77
View File
@@ -54,6 +54,60 @@ namespace MobileGL::MG_Pipe {
namespace {
Uint64 g_residualPulls = 0;
#if MOBILEGL_PIPE_VERIFY
// ---- P7 wave 3 (V1): negative control B's SERVER half ----------------------------
//
// MOBILEGL_PIPE_POISON_OMIT withholds the STAMP of one (verb, field) pair, and until a
// verify build first ran on the split arm it was only ever read by the client's filler
// (MG_Impl/Pipe/PipeFill.cpp's ParsePoisonOmissionKnob). On the split arm that is not
// where the stamp the backend reads comes from: the stamp below re-stamps every field
// the verb's class can answer out of applied records, so it restamped the very field
// the client had withheld and the control went GREEN - measured, both backends, the
// child's glGenerateMipmap returned and the process exited 0 with
// GenerateMipmap:GetActiveTextureUnit omitted. The knob means "this verb's stamp of this
// field was never written", and under a transport the stamp that answers the backend is
// this one, so this one honours it too.
//
// PARSED HERE, NOT SHARED WITH THE CLIENT'S PARSER, because this file is in the server
// library and MG_Impl is not. A malformed value is the client's Fatal{PipeVerifyBadKnob}
// (it parses first, at the first fill of the process); this side simply does not arm.
// VERIFY BUILDS ONLY: the knob belongs to the verify harness, and a push build's stamp
// stays the branch-free fill the P5d profile asked for.
struct ServerPoisonOmission {
Bool Parsed = false;
String Value;
Bool Armed = false;
MGPipeVerb Verb = MGPipeVerb::kVerbCount;
MGPipeInputField Field = MGPipeInputField::kFieldCount;
};
ServerPoisonOmission g_serverOmission;
const ServerPoisonOmission& ServerOmission() {
const String& knob = MG_Config::Features.PipePoisonOmit;
if (g_serverOmission.Parsed && knob.size() == g_serverOmission.Value.size() &&
(knob.empty() || knob == g_serverOmission.Value)) {
return g_serverOmission;
}
g_serverOmission = ServerPoisonOmission{};
g_serverOmission.Parsed = true;
g_serverOmission.Value = knob;
const auto colon = knob.find(':');
if (colon == String::npos || colon == 0 || colon + 1 >= knob.size()) return g_serverOmission;
const auto verb = MGPipeFindVerb(knob.substr(0, colon).c_str());
const auto field = MGPipeFindInputField(knob.substr(colon + 1).c_str());
if (!verb || !field) return g_serverOmission;
g_serverOmission.Armed = true;
g_serverOmission.Verb = *verb;
g_serverOmission.Field = *field;
return g_serverOmission;
}
Bool ServerOmitsStamp(MGPipeVerb verb, MGPipeInputField field) {
const ServerPoisonOmission& omission = ServerOmission();
return omission.Armed && omission.Verb == verb && omission.Field == field;
}
#endif // MOBILEGL_PIPE_VERIFY
// The strict arm of R-7.3. Same first line as the ordinary poison Fatal, so every
// existing filter on Fatal{UnmigratedPipeInput still matches, plus the class and the
// phase that retires it - a strict abort that did not say which phase owes the answer
@@ -352,6 +406,14 @@ namespace MobileGL::MG_Pipe {
Bool MGPipeServerContextIsLive() { return g_serverContextLive; }
#if MOBILEGL_BUILD_DISAGGREGATED
namespace { const void* g_serverOwnedWindow = nullptr; }
void MGPipeServerSetOwnedWindow(const void* window) { g_serverOwnedWindow = window; }
const void* MGPipeServerOwnedWindow() { return g_serverOwnedWindow; }
#endif
void MGPipeServerBlockNoteIdentity() {
// D10: THE WIDER PREDICATE, and the narrow one was a latent crash rather than a missing
// optimisation. Gated on the REHEARSAL, this early-returned for the whole life of a
@@ -415,6 +477,15 @@ namespace MobileGL::MG_Pipe {
const Uint64 bit = (answerable.Words[i / 64] >> (i % 64)) & Uint64{1};
filled.FilledGen[i] = serial & (Uint64{0} - bit);
}
#if MOBILEGL_PIPE_VERIFY
// Negative control B's server half (ServerPoisonOmission above): the one withheld pair.
{
const ServerPoisonOmission& omission = ServerOmission();
if (omission.Armed && omission.Verb == verb) {
filled.FilledGen[static_cast<SizeT>(omission.Field)] = 0;
}
}
#endif
MGPipeStampAccess::SetServerStamped(inputs, true);
// P5f (f1): the server block's identity rides the stamp (CONTRACT-P5E §3.2's
// "SetIdentity moves to ApplyOne" - the stamp is the per-verb half of ApplyOne, and
@@ -438,6 +509,12 @@ namespace MobileGL::MG_Pipe {
// AND A READ OUTSIDE THE VERB'S OWN CLASS IS STILL FATAL even for a BARRIER-PULLED
// field: the value it would be answered with was never copied for this verb, so
// counting it would trade a loud staleness for a quiet one.
#if MOBILEGL_PIPE_VERIFY
// An injected omission is reported as what it models - a stamp nobody wrote, the
// client filler's own Fatal{UnmigratedPipeInput, "<Field>@<Verb>"} - and never as the
// RoleViolation below, which would blame the wire for a withdrawal the control made.
if (serverStamped && ServerOmitsStamp(verb, field)) MGPipeInputPoisonFatalForVerb(field, verb);
#endif
if (!serverStamped ||
kMGPipeFieldOwnership[static_cast<SizeT>(field)] != MGPipeFieldOwnership::kBarrierPulled ||
!FieldIsInVerbClass(field, verb)) {
+14
View File
@@ -938,6 +938,13 @@ namespace MobileGL::MG_Pipe {
// GLContext or resurrected by stamping a verb after teardown.
void MGPipeServerSetContextLive(Bool live);
Bool MGPipeServerContextIsLive();
// P12 review fix (size reports without the knob). The SERVER's own display window while this
// session holds its lease (ServerLoop's ServerOwned arm sets it, the lease's end clears it), else
// null. A backend whose window surface does not otherwise publish its extent (Espryt) publishes
// it only for this window: a window the client named itself keeps the format-only publish every
// earlier session relied on. Apply thread only.
void MGPipeServerSetOwnedWindow(const void* window);
const void* MGPipeServerOwnedWindow();
#else
// The push-without-transport build has one role and one block, so the fill side's spelling
// folds onto gPipeInputs and PipeFill.cpp reads identically in both build flavours. An
@@ -1050,5 +1057,12 @@ namespace MobileGL::MG_Pipe {
// dereferenced, the snapshot is only ever compared). Returns false for a forwarded field,
// which has nothing to corrupt.
Bool MGPipeApplyVerifyCorruption(PipeInputs& snapshot, MGPipeInputField field);
// PipeFill.cpp. Writes the MOBILEGL_PIPE_VERIFY_FATAL=0 summary line ("N divergence(s)
// survived") NOW, while the role's log file is still open. MobileGL::Destroy calls it right
// before MG_Util::Debug::Close(): the line used to come from a namespace-scope static's
// destructor, which runs AFTER Close, and Log.cpp reopens a closed sink with "w" - so that
// summary was the only line the client half of every FATAL=0 run kept. Idempotent: a second
// call with nothing new to report, the destructor's own included, writes nothing.
void MGPipeVerifyFlushSummary();
#endif
} // namespace MobileGL::MG_Pipe
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,900 @@
// MobileGL - MobileGL/MG_Client/generated_wire_trampoline.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
#include <Includes.h>
// GENERATED by scripts/generate_client_trampoline.py
MOBILEGL_GL_API void glActiveProgramEXT(GLuint program);
MOBILEGL_GL_API void glActiveShaderProgram(GLuint pipeline, GLuint program);
MOBILEGL_GL_API void glActiveTexture(GLenum texture);
MOBILEGL_GL_API void glAttachShader(GLuint program, GLuint shader);
MOBILEGL_GL_API void glBeginConditionalRender(GLuint id, GLenum mode);
MOBILEGL_GL_API void glBeginConditionalRenderNV(GLuint id, GLenum mode);
MOBILEGL_GL_API void glBeginPerfMonitorAMD(GLuint monitor);
MOBILEGL_GL_API void glBeginPerfQueryINTEL(GLuint queryHandle);
MOBILEGL_GL_API void glBeginQuery(GLenum target, GLuint id);
MOBILEGL_GL_API void glBeginQueryIndexed(GLenum target, GLuint index, GLuint id);
MOBILEGL_GL_API void glBeginTransformFeedback(GLenum primitiveMode);
MOBILEGL_GL_API void glBindAttribLocation(GLuint program, GLuint index, const GLchar* name);
MOBILEGL_GL_API void glBindBuffer(GLenum target, GLuint buffer);
MOBILEGL_GL_API void glBindBufferBase(GLenum target, GLuint index, GLuint buffer);
MOBILEGL_GL_API void glBindBufferRange(GLenum target, GLuint index, GLuint buffer, GLintptr offset, GLsizeiptr size);
MOBILEGL_GL_API void glBindBuffersBase(GLenum target, GLuint first, GLsizei count, const GLuint* buffers);
MOBILEGL_GL_API void glBindFragDataLocation(GLuint program, GLuint color, const GLchar* name);
MOBILEGL_GL_API void glBindFragDataLocationIndexed(GLuint program, GLuint colorNumber, GLuint index, const GLchar* name);
MOBILEGL_GL_API void glBindFramebuffer(GLenum target, GLuint framebuffer);
MOBILEGL_GL_API void glBindImageTexture(GLuint unit, GLuint texture, GLint level, GLboolean layered, GLint layer, GLenum access, GLenum format);
MOBILEGL_GL_API void glBindImageTextures(GLuint first, GLsizei count, const GLuint* textures);
MOBILEGL_GL_API void glBindMultiTextureEXT(GLenum texunit, GLenum target, GLuint texture);
MOBILEGL_GL_API void glBindProgramPipeline(GLuint pipeline);
MOBILEGL_GL_API void glBindRenderbuffer(GLenum target, GLuint renderbuffer);
MOBILEGL_GL_API void glBindSampler(GLuint unit, GLuint sampler);
MOBILEGL_GL_API void glBindSamplers(GLuint first, GLsizei count, const GLuint* samplers);
MOBILEGL_GL_API void glBindShadingRateImageNV(GLuint texture);
MOBILEGL_GL_API void glBindTexture(GLenum target, GLuint texture);
MOBILEGL_GL_API void glBindTextureUnit(GLuint unit, GLuint texture);
MOBILEGL_GL_API void glBindTextures(GLuint first, GLsizei count, const GLuint* textures);
MOBILEGL_GL_API void glBindTransformFeedback(GLenum target, GLuint id);
MOBILEGL_GL_API void glBindVertexArray(GLuint array);
MOBILEGL_GL_API void glBindVertexBuffer(GLuint bindingindex, GLuint buffer, GLintptr offset, GLsizei stride);
MOBILEGL_GL_API void glBlendColor(GLfloat red, GLfloat green, GLfloat blue, GLfloat alpha);
MOBILEGL_GL_API void glBlendEquation(GLenum mode);
MOBILEGL_GL_API void glBlendEquationSeparate(GLenum modeRGB, GLenum modeAlpha);
MOBILEGL_GL_API void glBlendEquationSeparatei(GLuint buf, GLenum modeRGB, GLenum modeAlpha);
MOBILEGL_GL_API void glBlendEquationSeparateiARB(GLuint buf, GLenum modeRGB, GLenum modeAlpha);
MOBILEGL_GL_API void glBlendEquationi(GLuint buf, GLenum mode);
MOBILEGL_GL_API void glBlendEquationiARB(GLuint buf, GLenum mode);
MOBILEGL_GL_API void glBlendFunc(GLenum sfactor, GLenum dfactor);
MOBILEGL_GL_API void glBlendFuncSeparate(GLenum sfactorRGB, GLenum dfactorRGB, GLenum sfactorAlpha, GLenum dfactorAlpha);
MOBILEGL_GL_API void glBlendFuncSeparatei(GLuint buf, GLenum srcRGB, GLenum dstRGB, GLenum srcAlpha, GLenum dstAlpha);
MOBILEGL_GL_API void glBlendFuncSeparateiARB(GLuint buf, GLenum srcRGB, GLenum dstRGB, GLenum srcAlpha, GLenum dstAlpha);
MOBILEGL_GL_API void glBlendFunci(GLuint buf, GLenum src, GLenum dst);
MOBILEGL_GL_API void glBlendFunciARB(GLuint buf, GLenum src, GLenum dst);
MOBILEGL_GL_API void glBlendParameteriNV(GLenum pname, GLint value);
MOBILEGL_GL_API void glBlitFramebuffer(GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, GLbitfield mask, GLenum filter);
MOBILEGL_GL_API void glBlitNamedFramebuffer(GLuint readFramebuffer, GLuint drawFramebuffer, GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, GLbitfield mask, GLenum filter);
MOBILEGL_GL_API void glBufferAddressRangeNV(GLenum pname, GLuint index, GLuint64EXT address, GLsizeiptr length);
MOBILEGL_GL_API void glBufferAttachMemoryNV(GLenum target, GLuint memory, GLuint64 offset);
MOBILEGL_GL_API void glBufferData(GLenum target, GLsizeiptr size, const void* data, GLenum usage);
MOBILEGL_GL_API void glBufferPageCommitmentARB(GLenum target, GLintptr offset, GLsizeiptr size, GLboolean commit);
MOBILEGL_GL_API void glBufferPageCommitmentMemNV(GLenum target, GLintptr offset, GLsizeiptr size, GLuint memory, GLuint64 memOffset, GLboolean commit);
MOBILEGL_GL_API void glBufferStorage(GLenum target, GLsizeiptr size, const void* data, GLbitfield flags);
MOBILEGL_GL_API void glBufferSubData(GLenum target, GLintptr offset, GLsizeiptr size, const void* data);
MOBILEGL_GL_API void glCallCommandListNV(GLuint list);
MOBILEGL_GL_API GLenum glCheckFramebufferStatus(GLenum target);
MOBILEGL_GL_API GLenum glCheckNamedFramebufferStatus(GLuint framebuffer, GLenum target);
MOBILEGL_GL_API GLenum glCheckNamedFramebufferStatusEXT(GLuint framebuffer, GLenum target);
MOBILEGL_GL_API void glClampColor(GLenum target, GLenum clamp);
MOBILEGL_GL_API void glClear(GLbitfield mask);
MOBILEGL_GL_API void glClearBufferSubData(GLenum target, GLenum internalformat, GLintptr offset, GLsizeiptr size, GLenum format, GLenum type, const void* data);
MOBILEGL_GL_API void glClearBufferfi(GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil);
MOBILEGL_GL_API void glClearColor(GLfloat red, GLfloat green, GLfloat blue, GLfloat alpha);
MOBILEGL_GL_API void glClearDepth(GLclampd depth);
MOBILEGL_GL_API void glClearDepthdNV(GLdouble depth);
MOBILEGL_GL_API void glClearDepthf(GLfloat d);
MOBILEGL_GL_API void glClearNamedBufferSubData(GLuint buffer, GLenum internalformat, GLintptr offset, GLsizeiptr size, GLenum format, GLenum type, const void* data);
MOBILEGL_GL_API void glClearNamedBufferSubDataEXT(GLuint buffer, GLenum internalformat, GLsizeiptr offset, GLsizeiptr size, GLenum format, GLenum type, const void* data);
MOBILEGL_GL_API void glClearNamedFramebufferfi(GLuint framebuffer, GLenum buffer, GLint drawbuffer, GLfloat depth, GLint stencil);
MOBILEGL_GL_API void glClearStencil(GLint s);
MOBILEGL_GL_API void glClientAttribDefaultEXT(GLbitfield mask);
MOBILEGL_GL_API GLenum glClientWaitSync(GLsync sync, GLbitfield flags, GLuint64 timeout);
MOBILEGL_GL_API void glClipControl(GLenum origin, GLenum depth);
MOBILEGL_GL_API void glColorFormatNV(GLint size, GLenum type, GLsizei stride);
MOBILEGL_GL_API void glColorMask(GLboolean red, GLboolean green, GLboolean blue, GLboolean alpha);
MOBILEGL_GL_API void glColorMaski(GLuint index, GLboolean r, GLboolean g, GLboolean b, GLboolean a);
MOBILEGL_GL_API void glCommandListSegmentsNV(GLuint list, GLuint segments);
MOBILEGL_GL_API void glCompileCommandListNV(GLuint list);
MOBILEGL_GL_API void glCompileShader(GLuint shader);
MOBILEGL_GL_API void glConservativeRasterParameterfNV(GLenum pname, GLfloat value);
MOBILEGL_GL_API void glConservativeRasterParameteriNV(GLenum pname, GLint param);
MOBILEGL_GL_API void glCopyBufferSubData(GLenum readTarget, GLenum writeTarget, GLintptr readOffset, GLintptr writeOffset, GLsizeiptr size);
MOBILEGL_GL_API void glCopyImageSubData(GLuint srcName, GLenum srcTarget, GLint srcLevel, GLint srcX, GLint srcY, GLint srcZ, GLuint dstName, GLenum dstTarget, GLint dstLevel, GLint dstX, GLint dstY, GLint dstZ, GLsizei srcWidth, GLsizei srcHeight, GLsizei srcDepth);
MOBILEGL_GL_API void glCopyMultiTexImage1DEXT(GLenum texunit, GLenum target, GLint level, GLenum internalformat, GLint x, GLint y, GLsizei width, GLint border);
MOBILEGL_GL_API void glCopyMultiTexImage2DEXT(GLenum texunit, GLenum target, GLint level, GLenum internalformat, GLint x, GLint y, GLsizei width, GLsizei height, GLint border);
MOBILEGL_GL_API void glCopyMultiTexSubImage1DEXT(GLenum texunit, GLenum target, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width);
MOBILEGL_GL_API void glCopyMultiTexSubImage2DEXT(GLenum texunit, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCopyMultiTexSubImage3DEXT(GLenum texunit, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCopyNamedBufferSubData(GLuint readBuffer, GLuint writeBuffer, GLintptr readOffset, GLintptr writeOffset, GLsizeiptr size);
MOBILEGL_GL_API void glCopyPathNV(GLuint resultPath, GLuint srcPath);
MOBILEGL_GL_API void glCopyTexImage1D(GLenum target, GLint level, GLenum internalformat, GLint x, GLint y, GLsizei width, GLint border);
MOBILEGL_GL_API void glCopyTexImage2D(GLenum target, GLint level, GLenum internalformat, GLint x, GLint y, GLsizei width, GLsizei height, GLint border);
MOBILEGL_GL_API void glCopyTexSubImage1D(GLenum target, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width);
MOBILEGL_GL_API void glCopyTexSubImage2D(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCopyTexSubImage3D(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCopyTextureImage1DEXT(GLuint texture, GLenum target, GLint level, GLenum internalformat, GLint x, GLint y, GLsizei width, GLint border);
MOBILEGL_GL_API void glCopyTextureImage2DEXT(GLuint texture, GLenum target, GLint level, GLenum internalformat, GLint x, GLint y, GLsizei width, GLsizei height, GLint border);
MOBILEGL_GL_API void glCopyTextureSubImage1D(GLuint texture, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width);
MOBILEGL_GL_API void glCopyTextureSubImage1DEXT(GLuint texture, GLenum target, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width);
MOBILEGL_GL_API void glCopyTextureSubImage2D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCopyTextureSubImage2DEXT(GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCopyTextureSubImage3D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCopyTextureSubImage3DEXT(GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glCoverFillPathNV(GLuint path, GLenum coverMode);
MOBILEGL_GL_API void glCoverStrokePathNV(GLuint path, GLenum coverMode);
MOBILEGL_GL_API void glCoverageModulationNV(GLenum components);
MOBILEGL_GL_API void glCoverageModulationTableNV(GLsizei n, const GLfloat* v);
MOBILEGL_GL_API void glCreateBuffers(GLsizei n, GLuint* buffers);
MOBILEGL_GL_API void glCreateCommandListsNV(GLsizei n, GLuint* lists);
MOBILEGL_GL_API void glCreateFramebuffers(GLsizei n, GLuint* framebuffers);
MOBILEGL_GL_API void glCreateProgramPipelines(GLsizei n, GLuint* pipelines);
MOBILEGL_GL_API void glCreateQueries(GLenum target, GLsizei n, GLuint* ids);
MOBILEGL_GL_API void glCreateRenderbuffers(GLsizei n, GLuint* renderbuffers);
MOBILEGL_GL_API void glCreateSamplers(GLsizei n, GLuint* samplers);
MOBILEGL_GL_API GLuint glCreateShader(GLenum type);
MOBILEGL_GL_API GLuint glCreateShaderProgramEXT(GLenum type, const GLchar* string);
MOBILEGL_GL_API void glCreateStatesNV(GLsizei n, GLuint* states);
MOBILEGL_GL_API void glCreateTextures(GLenum target, GLsizei n, GLuint* textures);
MOBILEGL_GL_API void glCreateTransformFeedbacks(GLsizei n, GLuint* ids);
MOBILEGL_GL_API void glCreateVertexArrays(GLsizei n, GLuint* arrays);
MOBILEGL_GL_API void glCullFace(GLenum mode);
MOBILEGL_GL_API void glDebugMessageControl(GLenum source, GLenum type, GLenum severity, GLsizei count, const GLuint* ids, GLboolean enabled);
MOBILEGL_GL_API void glDebugMessageInsert(GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length, const GLchar* buf);
MOBILEGL_GL_API void glDeleteBuffers(GLsizei n, const GLuint* buffers);
MOBILEGL_GL_API void glDeleteCommandListsNV(GLsizei n, const GLuint* lists);
MOBILEGL_GL_API void glDeleteFramebuffers(GLsizei n, const GLuint* framebuffers);
MOBILEGL_GL_API void glDeleteNamedStringARB(GLint namelen, const GLchar* name);
MOBILEGL_GL_API void glDeletePathsNV(GLuint path, GLsizei range);
MOBILEGL_GL_API void glDeletePerfMonitorsAMD(GLsizei n, GLuint* monitors);
MOBILEGL_GL_API void glDeletePerfQueryINTEL(GLuint queryHandle);
MOBILEGL_GL_API void glDeleteProgram(GLuint program);
MOBILEGL_GL_API void glDeleteProgramPipelines(GLsizei n, const GLuint* pipelines);
MOBILEGL_GL_API void glDeleteQueries(GLsizei n, const GLuint* ids);
MOBILEGL_GL_API void glDeleteRenderbuffers(GLsizei n, const GLuint* renderbuffers);
MOBILEGL_GL_API void glDeleteSamplers(GLsizei count, const GLuint* samplers);
MOBILEGL_GL_API void glDeleteShader(GLuint shader);
MOBILEGL_GL_API void glDeleteStatesNV(GLsizei n, const GLuint* states);
MOBILEGL_GL_API void glDeleteSync(GLsync sync);
MOBILEGL_GL_API void glDeleteTextures(GLsizei n, const GLuint* textures);
MOBILEGL_GL_API void glDeleteTransformFeedbacks(GLsizei n, const GLuint* ids);
MOBILEGL_GL_API void glDeleteVertexArrays(GLsizei n, const GLuint* arrays);
MOBILEGL_GL_API void glDepthBoundsdNV(GLdouble zmin, GLdouble zmax);
MOBILEGL_GL_API void glDepthFunc(GLenum func);
MOBILEGL_GL_API void glDepthMask(GLboolean flag);
MOBILEGL_GL_API void glDepthRange(GLclampd near_val, GLclampd far_val);
MOBILEGL_GL_API void glDepthRangeArraydvNV(GLuint first, GLsizei count, const GLdouble* v);
MOBILEGL_GL_API void glDepthRangeArrayv(GLuint first, GLsizei count, const GLdouble* v);
MOBILEGL_GL_API void glDepthRangeIndexed(GLuint index, GLdouble n, GLdouble f);
MOBILEGL_GL_API void glDepthRangeIndexeddNV(GLuint index, GLdouble n, GLdouble f);
MOBILEGL_GL_API void glDepthRangedNV(GLdouble zNear, GLdouble zFar);
MOBILEGL_GL_API void glDepthRangef(GLfloat n, GLfloat f);
MOBILEGL_GL_API void glDetachShader(GLuint program, GLuint shader);
MOBILEGL_GL_API void glDisable(GLenum cap);
MOBILEGL_GL_API void glDisableClientStateIndexedEXT(GLenum array, GLuint index);
MOBILEGL_GL_API void glDisableClientStateiEXT(GLenum array, GLuint index);
MOBILEGL_GL_API void glDisableIndexedEXT(GLenum target, GLuint index);
MOBILEGL_GL_API void glDisableVertexArrayAttrib(GLuint vaobj, GLuint index);
MOBILEGL_GL_API void glDisableVertexArrayAttribEXT(GLuint vaobj, GLuint index);
MOBILEGL_GL_API void glDisableVertexArrayEXT(GLuint vaobj, GLenum array);
MOBILEGL_GL_API void glDisableVertexAttribArray(GLuint index);
MOBILEGL_GL_API void glDisablei(GLenum target, GLuint index);
MOBILEGL_GL_API void glDispatchCompute(GLuint num_groups_x, GLuint num_groups_y, GLuint num_groups_z);
MOBILEGL_GL_API void glDispatchComputeGroupSizeARB(GLuint num_groups_x, GLuint num_groups_y, GLuint num_groups_z, GLuint group_size_x, GLuint group_size_y, GLuint group_size_z);
MOBILEGL_GL_API void glDispatchComputeIndirect(GLintptr indirect);
MOBILEGL_GL_API void glDrawArrays(GLenum mode, GLint first, GLsizei count);
MOBILEGL_GL_API void glDrawArraysInstanced(GLenum mode, GLint first, GLsizei count, GLsizei instancecount);
MOBILEGL_GL_API void glDrawArraysInstancedBaseInstance(GLenum mode, GLint first, GLsizei count, GLsizei instancecount, GLuint baseinstance);
MOBILEGL_GL_API void glDrawBuffer(GLenum mode);
MOBILEGL_GL_API void glDrawBuffers(GLsizei n, const GLenum* bufs);
MOBILEGL_GL_API void glDrawCommandsAddressNV(GLenum primitiveMode, const GLuint64* indirects, const GLsizei* sizes, GLuint count);
MOBILEGL_GL_API void glDrawCommandsStatesAddressNV(const GLuint64* indirects, const GLsizei* sizes, const GLuint* states, const GLuint* fbos, GLuint count);
MOBILEGL_GL_API void glDrawElements(GLenum mode, GLsizei count, GLenum type, const void* indices);
MOBILEGL_GL_API void glDrawElementsBaseVertex(GLenum mode, GLsizei count, GLenum type, const void* indices, GLint basevertex);
MOBILEGL_GL_API void glDrawElementsInstanced(GLenum mode, GLsizei count, GLenum type, const void* indices, GLsizei instancecount);
MOBILEGL_GL_API void glDrawElementsInstancedBaseInstance(GLenum mode, GLsizei count, GLenum type, const void* indices, GLsizei instancecount, GLuint baseinstance);
MOBILEGL_GL_API void glDrawElementsInstancedBaseVertex(GLenum mode, GLsizei count, GLenum type, const void* indices, GLsizei instancecount, GLint basevertex);
MOBILEGL_GL_API void glDrawElementsInstancedBaseVertexBaseInstance(GLenum mode, GLsizei count, GLenum type, const void* indices, GLsizei instancecount, GLint basevertex, GLuint baseinstance);
MOBILEGL_GL_API void glDrawMeshTasksIndirectNV(GLintptr indirect);
MOBILEGL_GL_API void glDrawMeshTasksNV(GLuint first, GLuint count);
MOBILEGL_GL_API void glDrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type, const void* indices);
MOBILEGL_GL_API void glDrawRangeElementsBaseVertex(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type, const void* indices, GLint basevertex);
MOBILEGL_GL_API void glDrawTransformFeedback(GLenum mode, GLuint id);
MOBILEGL_GL_API void glDrawTransformFeedbackInstanced(GLenum mode, GLuint id, GLsizei instancecount);
MOBILEGL_GL_API void glDrawTransformFeedbackStream(GLenum mode, GLuint id, GLuint stream);
MOBILEGL_GL_API void glDrawTransformFeedbackStreamInstanced(GLenum mode, GLuint id, GLuint stream, GLsizei instancecount);
MOBILEGL_GL_API void glDrawVkImageNV(GLuint64 vkImage, GLuint sampler, GLfloat x0, GLfloat y0, GLfloat x1, GLfloat y1, GLfloat z, GLfloat s0, GLfloat t0, GLfloat s1, GLfloat t1);
MOBILEGL_GL_API void glEdgeFlagFormatNV(GLsizei stride);
MOBILEGL_GL_API void glEnable(GLenum cap);
MOBILEGL_GL_API void glEnableClientStateIndexedEXT(GLenum array, GLuint index);
MOBILEGL_GL_API void glEnableClientStateiEXT(GLenum array, GLuint index);
MOBILEGL_GL_API void glEnableIndexedEXT(GLenum target, GLuint index);
MOBILEGL_GL_API void glEnableVertexArrayAttrib(GLuint vaobj, GLuint index);
MOBILEGL_GL_API void glEnableVertexArrayAttribEXT(GLuint vaobj, GLuint index);
MOBILEGL_GL_API void glEnableVertexArrayEXT(GLuint vaobj, GLenum array);
MOBILEGL_GL_API void glEnableVertexAttribArray(GLuint index);
MOBILEGL_GL_API void glEnablei(GLenum target, GLuint index);
MOBILEGL_GL_API void glEndPerfMonitorAMD(GLuint monitor);
MOBILEGL_GL_API void glEndPerfQueryINTEL(GLuint queryHandle);
MOBILEGL_GL_API void glEndQuery(GLenum target);
MOBILEGL_GL_API void glEndQueryIndexed(GLenum target, GLuint index);
MOBILEGL_GL_API void glFlushMappedBufferRange(GLenum target, GLintptr offset, GLsizeiptr length);
MOBILEGL_GL_API void glFlushMappedNamedBufferRange(GLuint buffer, GLintptr offset, GLsizeiptr length);
MOBILEGL_GL_API void glFlushMappedNamedBufferRangeEXT(GLuint buffer, GLintptr offset, GLsizeiptr length);
MOBILEGL_GL_API void glFogCoordFormatNV(GLenum type, GLsizei stride);
MOBILEGL_GL_API void glFragmentCoverageColorNV(GLuint color);
MOBILEGL_GL_API void glFramebufferDrawBufferEXT(GLuint framebuffer, GLenum mode);
MOBILEGL_GL_API void glFramebufferDrawBuffersEXT(GLuint framebuffer, GLsizei n, const GLenum* bufs);
MOBILEGL_GL_API void glFramebufferParameteri(GLenum target, GLenum pname, GLint param);
MOBILEGL_GL_API void glFramebufferParameteriMESA(GLenum target, GLenum pname, GLint param);
MOBILEGL_GL_API void glFramebufferReadBufferEXT(GLuint framebuffer, GLenum mode);
MOBILEGL_GL_API void glFramebufferRenderbuffer(GLenum target, GLenum attachment, GLenum renderbuffertarget, GLuint renderbuffer);
MOBILEGL_GL_API void glFramebufferSampleLocationsfvARB(GLenum target, GLuint start, GLsizei count, const GLfloat* v);
MOBILEGL_GL_API void glFramebufferSampleLocationsfvNV(GLenum target, GLuint start, GLsizei count, const GLfloat* v);
MOBILEGL_GL_API void glFramebufferTexture(GLenum target, GLenum attachment, GLuint texture, GLint level);
MOBILEGL_GL_API void glFramebufferTexture1D(GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level);
MOBILEGL_GL_API void glFramebufferTexture1DEXT(GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level);
MOBILEGL_GL_API void glFramebufferTexture2D(GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level);
MOBILEGL_GL_API void glFramebufferTexture3D(GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level, GLint zoffset);
MOBILEGL_GL_API void glFramebufferTexture3DEXT(GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level, GLint zoffset);
MOBILEGL_GL_API void glFramebufferTextureFaceARB(GLenum target, GLenum attachment, GLuint texture, GLint level, GLenum face);
MOBILEGL_GL_API void glFramebufferTextureLayer(GLenum target, GLenum attachment, GLuint texture, GLint level, GLint layer);
MOBILEGL_GL_API void glFramebufferTextureMultiviewOVR(GLenum target, GLenum attachment, GLuint texture, GLint level, GLint baseViewIndex, GLsizei numViews);
MOBILEGL_GL_API void glFrontFace(GLenum mode);
MOBILEGL_GL_API void glGenBuffers(GLsizei n, GLuint* buffers);
MOBILEGL_GL_API void glGenFramebuffers(GLsizei n, GLuint* framebuffers);
MOBILEGL_GL_API GLuint glGenPathsNV(GLsizei range);
MOBILEGL_GL_API void glGenPerfMonitorsAMD(GLsizei n, GLuint* monitors);
MOBILEGL_GL_API void glGenProgramPipelines(GLsizei n, GLuint* pipelines);
MOBILEGL_GL_API void glGenQueries(GLsizei n, GLuint* ids);
MOBILEGL_GL_API void glGenRenderbuffers(GLsizei n, GLuint* renderbuffers);
MOBILEGL_GL_API void glGenSamplers(GLsizei count, GLuint* samplers);
MOBILEGL_GL_API void glGenTextures(GLsizei n, GLuint* textures);
MOBILEGL_GL_API void glGenTransformFeedbacks(GLsizei n, GLuint* ids);
MOBILEGL_GL_API void glGenVertexArrays(GLsizei n, GLuint* arrays);
MOBILEGL_GL_API void glGenerateMipmap(GLenum target);
MOBILEGL_GL_API void glGenerateMultiTexMipmapEXT(GLenum texunit, GLenum target);
MOBILEGL_GL_API void glGenerateTextureMipmap(GLuint texture);
MOBILEGL_GL_API void glGenerateTextureMipmapEXT(GLuint texture, GLenum target);
MOBILEGL_GL_API void glGetActiveAtomicCounterBufferiv(GLuint program, GLuint bufferIndex, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetActiveSubroutineUniformiv(GLuint program, GLenum shadertype, GLuint index, GLenum pname, GLint* values);
MOBILEGL_GL_API void glGetActiveUniformBlockiv(GLuint program, GLuint uniformBlockIndex, GLenum pname, GLint* params);
MOBILEGL_GL_API GLint glGetAttribLocation(GLuint program, const GLchar* name);
MOBILEGL_GL_API void glGetBooleanv(GLenum pname, GLboolean* data);
MOBILEGL_GL_API void glGetBufferParameteri64v(GLenum target, GLenum pname, GLint64* params);
MOBILEGL_GL_API void glGetBufferParameteriv(GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetBufferParameterui64vNV(GLenum target, GLenum pname, GLuint64EXT* params);
MOBILEGL_GL_API void glGetBufferSubData(GLenum target, GLintptr offset, GLsizeiptr size, void* data);
MOBILEGL_GL_API GLuint glGetCommandHeaderNV(GLenum tokenID, GLuint size);
MOBILEGL_GL_API GLuint glGetDebugMessageLog(GLuint count, GLsizei bufSize, GLenum* sources, GLenum* types, GLuint* ids, GLenum* severities, GLsizei* lengths, GLchar* messageLog);
MOBILEGL_GL_API void glGetDoublei_vEXT(GLenum pname, GLuint index, GLdouble* params);
MOBILEGL_GL_API void glGetDoublev(GLenum pname, GLdouble* params);
MOBILEGL_GL_API void glGetFloati_vEXT(GLenum pname, GLuint index, GLfloat* params);
MOBILEGL_GL_API void glGetFloatv(GLenum pname, GLfloat* data);
MOBILEGL_GL_API GLint glGetFragDataIndex(GLuint program, const GLchar* name);
MOBILEGL_GL_API GLint glGetFragDataLocation(GLuint program, const GLchar* name);
MOBILEGL_GL_API void glGetFramebufferAttachmentParameteriv(GLenum target, GLenum attachment, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetFramebufferParameteriv(GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetFramebufferParameterivMESA(GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API GLuint64 glGetImageHandleARB(GLuint texture, GLint level, GLboolean layered, GLint layer, GLenum format);
MOBILEGL_GL_API GLuint64 glGetImageHandleNV(GLuint texture, GLint level, GLboolean layered, GLint layer, GLenum format);
MOBILEGL_GL_API void glGetInteger64v(GLenum pname, GLint64* data);
MOBILEGL_GL_API void glGetIntegerv(GLenum pname, GLint* data);
MOBILEGL_GL_API void glGetInternalformatSampleivNV(GLenum target, GLenum internalformat, GLsizei samples, GLenum pname, GLsizei count, GLint* params);
MOBILEGL_GL_API void glGetInternalformati64v(GLenum target, GLenum internalformat, GLenum pname, GLsizei count, GLint64* params);
MOBILEGL_GL_API void glGetInternalformativ(GLenum target, GLenum internalformat, GLenum pname, GLsizei bufSize, GLint* params);
MOBILEGL_GL_API void glGetMemoryObjectDetachedResourcesuivNV(GLuint memory, GLenum pname, GLint first, GLsizei count, GLuint* params);
MOBILEGL_GL_API void glGetMultiTexEnvfvEXT(GLenum texunit, GLenum target, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetMultiTexEnvivEXT(GLenum texunit, GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetMultiTexGendvEXT(GLenum texunit, GLenum coord, GLenum pname, GLdouble* params);
MOBILEGL_GL_API void glGetMultiTexGenfvEXT(GLenum texunit, GLenum coord, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetMultiTexGenivEXT(GLenum texunit, GLenum coord, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetMultiTexLevelParameterfvEXT(GLenum texunit, GLenum target, GLint level, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetMultiTexLevelParameterivEXT(GLenum texunit, GLenum target, GLint level, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetMultiTexParameterIivEXT(GLenum texunit, GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetMultiTexParameterIuivEXT(GLenum texunit, GLenum target, GLenum pname, GLuint* params);
MOBILEGL_GL_API void glGetMultiTexParameterfvEXT(GLenum texunit, GLenum target, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetMultiTexParameterivEXT(GLenum texunit, GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetMultisamplefv(GLenum pname, GLuint index, GLfloat* val);
MOBILEGL_GL_API void glGetNamedBufferParameteri64v(GLuint buffer, GLenum pname, GLint64* params);
MOBILEGL_GL_API void glGetNamedBufferParameteriv(GLuint buffer, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedBufferParameterivEXT(GLuint buffer, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedBufferParameterui64vNV(GLuint buffer, GLenum pname, GLuint64EXT* params);
MOBILEGL_GL_API void glGetNamedBufferSubData(GLuint buffer, GLintptr offset, GLsizeiptr size, void* data);
MOBILEGL_GL_API void glGetNamedBufferSubDataEXT(GLuint buffer, GLintptr offset, GLsizeiptr size, void* data);
MOBILEGL_GL_API void glGetNamedFramebufferAttachmentParameteriv(GLuint framebuffer, GLenum attachment, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedFramebufferAttachmentParameterivEXT(GLuint framebuffer, GLenum attachment, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedFramebufferParameteriv(GLuint framebuffer, GLenum pname, GLint* param);
MOBILEGL_GL_API void glGetNamedFramebufferParameterivEXT(GLuint framebuffer, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedProgramivEXT(GLuint program, GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedRenderbufferParameteriv(GLuint renderbuffer, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedRenderbufferParameterivEXT(GLuint renderbuffer, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetNamedStringivARB(GLint namelen, const GLchar* name, GLenum pname, GLint* params);
MOBILEGL_GL_API GLfloat glGetPathLengthNV(GLuint path, GLsizei startSegment, GLsizei numSegments);
MOBILEGL_GL_API void glGetPathParameterfvNV(GLuint path, GLenum pname, GLfloat* value);
MOBILEGL_GL_API void glGetPathParameterivNV(GLuint path, GLenum pname, GLint* value);
MOBILEGL_GL_API void glGetPerfMonitorCounterDataAMD(GLuint monitor, GLenum pname, GLsizei dataSize, GLuint* data, GLint* bytesWritten);
MOBILEGL_GL_API void glGetProgramInterfaceiv(GLuint program, GLenum programInterface, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetProgramPipelineiv(GLuint pipeline, GLenum pname, GLint* params);
MOBILEGL_GL_API GLuint glGetProgramResourceIndex(GLuint program, GLenum programInterface, const GLchar* name);
MOBILEGL_GL_API GLint glGetProgramResourceLocation(GLuint program, GLenum programInterface, const GLchar* name);
MOBILEGL_GL_API GLint glGetProgramResourceLocationIndex(GLuint program, GLenum programInterface, const GLchar* name);
MOBILEGL_GL_API void glGetProgramResourcefvNV(GLuint program, GLenum programInterface, GLuint index, GLsizei propCount, const GLenum* props, GLsizei count, GLsizei* length, GLfloat* params);
MOBILEGL_GL_API void glGetProgramStageiv(GLuint program, GLenum shadertype, GLenum pname, GLint* values);
MOBILEGL_GL_API void glGetProgramiv(GLuint program, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetQueryBufferObjecti64v(GLuint id, GLuint buffer, GLenum pname, GLintptr offset);
MOBILEGL_GL_API void glGetQueryBufferObjectiv(GLuint id, GLuint buffer, GLenum pname, GLintptr offset);
MOBILEGL_GL_API void glGetQueryBufferObjectui64v(GLuint id, GLuint buffer, GLenum pname, GLintptr offset);
MOBILEGL_GL_API void glGetQueryBufferObjectuiv(GLuint id, GLuint buffer, GLenum pname, GLintptr offset);
MOBILEGL_GL_API void glGetQueryIndexediv(GLenum target, GLuint index, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetQueryObjecti64v(GLuint id, GLenum pname, GLint64* params);
MOBILEGL_GL_API void glGetQueryObjectiv(GLuint id, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetQueryObjectui64v(GLuint id, GLenum pname, GLuint64* params);
MOBILEGL_GL_API void glGetQueryObjectuiv(GLuint id, GLenum pname, GLuint* params);
MOBILEGL_GL_API void glGetQueryiv(GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetRenderbufferParameteriv(GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetSamplerParameterIiv(GLuint sampler, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetSamplerParameterIuiv(GLuint sampler, GLenum pname, GLuint* params);
MOBILEGL_GL_API void glGetSamplerParameterfv(GLuint sampler, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetSamplerParameteriv(GLuint sampler, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetShaderiv(GLuint shader, GLenum pname, GLint* params);
MOBILEGL_GL_API GLushort glGetStageIndexNV(GLenum shadertype);
MOBILEGL_GL_API const GLubyte* glGetString(GLenum name);
MOBILEGL_GL_API const GLubyte* glGetStringi(GLenum name, GLuint index);
MOBILEGL_GL_API GLuint glGetSubroutineIndex(GLuint program, GLenum shadertype, const GLchar* name);
MOBILEGL_GL_API GLint glGetSubroutineUniformLocation(GLuint program, GLenum shadertype, const GLchar* name);
MOBILEGL_GL_API void glGetSynciv(GLsync sync, GLenum pname, GLsizei bufSize, GLsizei* length, GLint* values);
MOBILEGL_GL_API void glGetTexLevelParameterfv(GLenum target, GLint level, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetTexLevelParameteriv(GLenum target, GLint level, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetTexParameterIiv(GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetTexParameterIuiv(GLenum target, GLenum pname, GLuint* params);
MOBILEGL_GL_API void glGetTexParameterfv(GLenum target, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetTexParameteriv(GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API GLuint64 glGetTextureHandleARB(GLuint texture);
MOBILEGL_GL_API GLuint64 glGetTextureHandleNV(GLuint texture);
MOBILEGL_GL_API void glGetTextureLevelParameterfv(GLuint texture, GLint level, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetTextureLevelParameterfvEXT(GLuint texture, GLenum target, GLint level, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetTextureLevelParameteriv(GLuint texture, GLint level, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetTextureLevelParameterivEXT(GLuint texture, GLenum target, GLint level, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetTextureParameterIiv(GLuint texture, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetTextureParameterIivEXT(GLuint texture, GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetTextureParameterIuiv(GLuint texture, GLenum pname, GLuint* params);
MOBILEGL_GL_API void glGetTextureParameterIuivEXT(GLuint texture, GLenum target, GLenum pname, GLuint* params);
MOBILEGL_GL_API void glGetTextureParameterfv(GLuint texture, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetTextureParameterfvEXT(GLuint texture, GLenum target, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetTextureParameteriv(GLuint texture, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetTextureParameterivEXT(GLuint texture, GLenum target, GLenum pname, GLint* params);
MOBILEGL_GL_API GLuint64 glGetTextureSamplerHandleARB(GLuint texture, GLuint sampler);
MOBILEGL_GL_API GLuint64 glGetTextureSamplerHandleNV(GLuint texture, GLuint sampler);
MOBILEGL_GL_API void glGetTransformFeedbacki64_v(GLuint xfb, GLenum pname, GLuint index, GLint64* param);
MOBILEGL_GL_API void glGetTransformFeedbacki_v(GLuint xfb, GLenum pname, GLuint index, GLint* param);
MOBILEGL_GL_API void glGetTransformFeedbackiv(GLuint xfb, GLenum pname, GLint* param);
MOBILEGL_GL_API GLuint glGetUniformBlockIndex(GLuint program, const GLchar* uniformBlockName);
MOBILEGL_GL_API GLint glGetUniformLocation(GLuint program, const GLchar* name);
MOBILEGL_GL_API void glGetVertexArrayIndexed64iv(GLuint vaobj, GLuint index, GLenum pname, GLint64* param);
MOBILEGL_GL_API void glGetVertexArrayIndexediv(GLuint vaobj, GLuint index, GLenum pname, GLint* param);
MOBILEGL_GL_API void glGetVertexArrayIntegeri_vEXT(GLuint vaobj, GLuint index, GLenum pname, GLint* param);
MOBILEGL_GL_API void glGetVertexArrayIntegervEXT(GLuint vaobj, GLenum pname, GLint* param);
MOBILEGL_GL_API void glGetVertexArrayiv(GLuint vaobj, GLenum pname, GLint* param);
MOBILEGL_GL_API void glGetVertexAttribIiv(GLuint index, GLenum pname, GLint* params);
MOBILEGL_GL_API void glGetVertexAttribIuiv(GLuint index, GLenum pname, GLuint* params);
MOBILEGL_GL_API void glGetVertexAttribLdv(GLuint index, GLenum pname, GLdouble* params);
MOBILEGL_GL_API void glGetVertexAttribLi64vNV(GLuint index, GLenum pname, GLint64EXT* params);
MOBILEGL_GL_API void glGetVertexAttribLui64vARB(GLuint index, GLenum pname, GLuint64EXT* params);
MOBILEGL_GL_API void glGetVertexAttribLui64vNV(GLuint index, GLenum pname, GLuint64EXT* params);
MOBILEGL_GL_API void glGetVertexAttribdv(GLuint index, GLenum pname, GLdouble* params);
MOBILEGL_GL_API void glGetVertexAttribfv(GLuint index, GLenum pname, GLfloat* params);
MOBILEGL_GL_API void glGetVertexAttribiv(GLuint index, GLenum pname, GLint* params);
MOBILEGL_GL_API void glHint(GLenum target, GLenum mode);
MOBILEGL_GL_API void glIndexFormatNV(GLenum type, GLsizei stride);
MOBILEGL_GL_API void glInsertEventMarkerEXT(GLsizei length, const GLchar* marker);
MOBILEGL_GL_API void glInterpolatePathsNV(GLuint resultPath, GLuint pathA, GLuint pathB, GLfloat weight);
MOBILEGL_GL_API void glInvalidateBufferData(GLuint buffer);
MOBILEGL_GL_API void glInvalidateBufferSubData(GLuint buffer, GLintptr offset, GLsizeiptr length);
MOBILEGL_GL_API void glInvalidateTexImage(GLuint texture, GLint level);
MOBILEGL_GL_API void glInvalidateTexSubImage(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth);
MOBILEGL_GL_API GLboolean glIsBuffer(GLuint buffer);
MOBILEGL_GL_API GLboolean glIsBufferResidentNV(GLenum target);
MOBILEGL_GL_API GLboolean glIsCommandListNV(GLuint list);
MOBILEGL_GL_API GLboolean glIsEnabled(GLenum cap);
MOBILEGL_GL_API GLboolean glIsEnabledIndexedEXT(GLenum target, GLuint index);
MOBILEGL_GL_API GLboolean glIsEnabledi(GLenum target, GLuint index);
MOBILEGL_GL_API GLboolean glIsFramebuffer(GLuint framebuffer);
MOBILEGL_GL_API GLboolean glIsImageHandleResidentARB(GLuint64 handle);
MOBILEGL_GL_API GLboolean glIsImageHandleResidentNV(GLuint64 handle);
MOBILEGL_GL_API GLboolean glIsNamedBufferResidentNV(GLuint buffer);
MOBILEGL_GL_API GLboolean glIsNamedStringARB(GLint namelen, const GLchar* name);
MOBILEGL_GL_API GLboolean glIsPathNV(GLuint path);
MOBILEGL_GL_API GLboolean glIsPointInFillPathNV(GLuint path, GLuint mask, GLfloat x, GLfloat y);
MOBILEGL_GL_API GLboolean glIsPointInStrokePathNV(GLuint path, GLfloat x, GLfloat y);
MOBILEGL_GL_API GLboolean glIsProgram(GLuint program);
MOBILEGL_GL_API GLboolean glIsProgramPipeline(GLuint pipeline);
MOBILEGL_GL_API GLboolean glIsQuery(GLuint id);
MOBILEGL_GL_API GLboolean glIsRenderbuffer(GLuint renderbuffer);
MOBILEGL_GL_API GLboolean glIsSampler(GLuint sampler);
MOBILEGL_GL_API GLboolean glIsShader(GLuint shader);
MOBILEGL_GL_API GLboolean glIsStateNV(GLuint state);
MOBILEGL_GL_API GLboolean glIsSync(GLsync sync);
MOBILEGL_GL_API GLboolean glIsTexture(GLuint texture);
MOBILEGL_GL_API GLboolean glIsTextureHandleResidentARB(GLuint64 handle);
MOBILEGL_GL_API GLboolean glIsTextureHandleResidentNV(GLuint64 handle);
MOBILEGL_GL_API GLboolean glIsTransformFeedback(GLuint id);
MOBILEGL_GL_API GLboolean glIsVertexArray(GLuint array);
MOBILEGL_GL_API void glLabelObjectEXT(GLenum type, GLuint object, GLsizei length, const GLchar* label);
MOBILEGL_GL_API void glLineWidth(GLfloat width);
MOBILEGL_GL_API void glLinkProgram(GLuint program);
MOBILEGL_GL_API void glLogicOp(GLenum opcode);
MOBILEGL_GL_API void glMakeBufferNonResidentNV(GLenum target);
MOBILEGL_GL_API void glMakeBufferResidentNV(GLenum target, GLenum access);
MOBILEGL_GL_API void glMakeImageHandleNonResidentARB(GLuint64 handle);
MOBILEGL_GL_API void glMakeImageHandleNonResidentNV(GLuint64 handle);
MOBILEGL_GL_API void glMakeImageHandleResidentARB(GLuint64 handle, GLenum access);
MOBILEGL_GL_API void glMakeImageHandleResidentNV(GLuint64 handle, GLenum access);
MOBILEGL_GL_API void glMakeNamedBufferNonResidentNV(GLuint buffer);
MOBILEGL_GL_API void glMakeNamedBufferResidentNV(GLuint buffer, GLenum access);
MOBILEGL_GL_API void glMakeTextureHandleNonResidentARB(GLuint64 handle);
MOBILEGL_GL_API void glMakeTextureHandleNonResidentNV(GLuint64 handle);
MOBILEGL_GL_API void glMakeTextureHandleResidentARB(GLuint64 handle);
MOBILEGL_GL_API void glMakeTextureHandleResidentNV(GLuint64 handle);
MOBILEGL_GL_API void* glMapBuffer(GLenum target, GLenum access);
MOBILEGL_GL_API void* glMapBufferRange(GLenum target, GLintptr offset, GLsizeiptr length, GLbitfield access);
MOBILEGL_GL_API void glMatrixFrustumEXT(GLenum mode, GLdouble left, GLdouble right, GLdouble bottom, GLdouble top, GLdouble zNear, GLdouble zFar);
MOBILEGL_GL_API void glMatrixLoadIdentityEXT(GLenum mode);
MOBILEGL_GL_API void glMatrixOrthoEXT(GLenum mode, GLdouble left, GLdouble right, GLdouble bottom, GLdouble top, GLdouble zNear, GLdouble zFar);
MOBILEGL_GL_API void glMatrixPopEXT(GLenum mode);
MOBILEGL_GL_API void glMatrixPushEXT(GLenum mode);
MOBILEGL_GL_API void glMatrixRotatedEXT(GLenum mode, GLdouble ane, GLdouble x, GLdouble y, GLdouble z);
MOBILEGL_GL_API void glMatrixRotatefEXT(GLenum mode, GLfloat ane, GLfloat x, GLfloat y, GLfloat z);
MOBILEGL_GL_API void glMatrixScaledEXT(GLenum mode, GLdouble x, GLdouble y, GLdouble z);
MOBILEGL_GL_API void glMatrixScalefEXT(GLenum mode, GLfloat x, GLfloat y, GLfloat z);
MOBILEGL_GL_API void glMatrixTranslatedEXT(GLenum mode, GLdouble x, GLdouble y, GLdouble z);
MOBILEGL_GL_API void glMatrixTranslatefEXT(GLenum mode, GLfloat x, GLfloat y, GLfloat z);
MOBILEGL_GL_API void glMaxShaderCompilerThreadsARB(GLuint count);
MOBILEGL_GL_API void glMaxShaderCompilerThreadsKHR(GLuint count);
MOBILEGL_GL_API void glMemoryBarrier(GLbitfield barriers);
MOBILEGL_GL_API void glMemoryBarrierByRegion(GLbitfield barriers);
MOBILEGL_GL_API void glMinSampleShading(GLfloat value);
MOBILEGL_GL_API void glMultiDrawMeshTasksIndirectCountNV(GLintptr indirect, GLintptr drawcount, GLsizei maxdrawcount, GLsizei stride);
MOBILEGL_GL_API void glMultiDrawMeshTasksIndirectNV(GLintptr indirect, GLsizei drawcount, GLsizei stride);
MOBILEGL_GL_API void glMultiTexBufferEXT(GLenum texunit, GLenum target, GLenum internalformat, GLuint buffer);
MOBILEGL_GL_API void glMultiTexCoordPointerEXT(GLenum texunit, GLint size, GLenum type, GLsizei stride, const void* pointer);
MOBILEGL_GL_API void glMultiTexEnvfEXT(GLenum texunit, GLenum target, GLenum pname, GLfloat param);
MOBILEGL_GL_API void glMultiTexEnviEXT(GLenum texunit, GLenum target, GLenum pname, GLint param);
MOBILEGL_GL_API void glMultiTexGendEXT(GLenum texunit, GLenum coord, GLenum pname, GLdouble param);
MOBILEGL_GL_API void glMultiTexGenfEXT(GLenum texunit, GLenum coord, GLenum pname, GLfloat param);
MOBILEGL_GL_API void glMultiTexGeniEXT(GLenum texunit, GLenum coord, GLenum pname, GLint param);
MOBILEGL_GL_API void glMultiTexParameterfEXT(GLenum texunit, GLenum target, GLenum pname, GLfloat param);
MOBILEGL_GL_API void glMultiTexParameteriEXT(GLenum texunit, GLenum target, GLenum pname, GLint param);
MOBILEGL_GL_API void glMultiTexRenderbufferEXT(GLenum texunit, GLenum target, GLuint renderbuffer);
MOBILEGL_GL_API void glNamedBufferAttachMemoryNV(GLuint buffer, GLuint memory, GLuint64 offset);
MOBILEGL_GL_API void glNamedBufferData(GLuint buffer, GLsizeiptr size, const void* data, GLenum usage);
MOBILEGL_GL_API void glNamedBufferDataEXT(GLuint buffer, GLsizeiptr size, const void* data, GLenum usage);
MOBILEGL_GL_API void glNamedBufferPageCommitmentARB(GLuint buffer, GLintptr offset, GLsizeiptr size, GLboolean commit);
MOBILEGL_GL_API void glNamedBufferPageCommitmentEXT(GLuint buffer, GLintptr offset, GLsizeiptr size, GLboolean commit);
MOBILEGL_GL_API void glNamedBufferPageCommitmentMemNV(GLuint buffer, GLintptr offset, GLsizeiptr size, GLuint memory, GLuint64 memOffset, GLboolean commit);
MOBILEGL_GL_API void glNamedBufferStorage(GLuint buffer, GLsizeiptr size, const void* data, GLbitfield flags);
MOBILEGL_GL_API void glNamedBufferStorageEXT(GLuint buffer, GLsizeiptr size, const void* data, GLbitfield flags);
MOBILEGL_GL_API void glNamedBufferSubData(GLuint buffer, GLintptr offset, GLsizeiptr size, const void* data);
MOBILEGL_GL_API void glNamedBufferSubDataEXT(GLuint buffer, GLintptr offset, GLsizeiptr size, const void* data);
MOBILEGL_GL_API void glNamedCopyBufferSubDataEXT(GLuint readBuffer, GLuint writeBuffer, GLintptr readOffset, GLintptr writeOffset, GLsizeiptr size);
MOBILEGL_GL_API void glNamedFramebufferDrawBuffer(GLuint framebuffer, GLenum buf);
MOBILEGL_GL_API void glNamedFramebufferDrawBuffers(GLuint framebuffer, GLsizei n, const GLenum* bufs);
MOBILEGL_GL_API void glNamedFramebufferParameteri(GLuint framebuffer, GLenum pname, GLint param);
MOBILEGL_GL_API void glNamedFramebufferParameteriEXT(GLuint framebuffer, GLenum pname, GLint param);
MOBILEGL_GL_API void glNamedFramebufferReadBuffer(GLuint framebuffer, GLenum src);
MOBILEGL_GL_API void glNamedFramebufferRenderbuffer(GLuint framebuffer, GLenum attachment, GLenum renderbuffertarget, GLuint renderbuffer);
MOBILEGL_GL_API void glNamedFramebufferRenderbufferEXT(GLuint framebuffer, GLenum attachment, GLenum renderbuffertarget, GLuint renderbuffer);
MOBILEGL_GL_API void glNamedFramebufferSampleLocationsfvARB(GLuint framebuffer, GLuint start, GLsizei count, const GLfloat* v);
MOBILEGL_GL_API void glNamedFramebufferSampleLocationsfvNV(GLuint framebuffer, GLuint start, GLsizei count, const GLfloat* v);
MOBILEGL_GL_API void glNamedFramebufferTexture(GLuint framebuffer, GLenum attachment, GLuint texture, GLint level);
MOBILEGL_GL_API void glNamedFramebufferTexture1DEXT(GLuint framebuffer, GLenum attachment, GLenum textarget, GLuint texture, GLint level);
MOBILEGL_GL_API void glNamedFramebufferTexture2DEXT(GLuint framebuffer, GLenum attachment, GLenum textarget, GLuint texture, GLint level);
MOBILEGL_GL_API void glNamedFramebufferTexture3DEXT(GLuint framebuffer, GLenum attachment, GLenum textarget, GLuint texture, GLint level, GLint zoffset);
MOBILEGL_GL_API void glNamedFramebufferTextureEXT(GLuint framebuffer, GLenum attachment, GLuint texture, GLint level);
MOBILEGL_GL_API void glNamedFramebufferTextureFaceEXT(GLuint framebuffer, GLenum attachment, GLuint texture, GLint level, GLenum face);
MOBILEGL_GL_API void glNamedFramebufferTextureLayer(GLuint framebuffer, GLenum attachment, GLuint texture, GLint level, GLint layer);
MOBILEGL_GL_API void glNamedFramebufferTextureLayerEXT(GLuint framebuffer, GLenum attachment, GLuint texture, GLint level, GLint layer);
MOBILEGL_GL_API void glNamedProgramLocalParameter4dEXT(GLuint program, GLenum target, GLuint index, GLdouble x, GLdouble y, GLdouble z, GLdouble w);
MOBILEGL_GL_API void glNamedProgramLocalParameter4fEXT(GLuint program, GLenum target, GLuint index, GLfloat x, GLfloat y, GLfloat z, GLfloat w);
MOBILEGL_GL_API void glNamedProgramLocalParameterI4iEXT(GLuint program, GLenum target, GLuint index, GLint x, GLint y, GLint z, GLint w);
MOBILEGL_GL_API void glNamedProgramLocalParameterI4uiEXT(GLuint program, GLenum target, GLuint index, GLuint x, GLuint y, GLuint z, GLuint w);
MOBILEGL_GL_API void glNamedProgramLocalParameters4fvEXT(GLuint program, GLenum target, GLuint index, GLsizei count, const GLfloat* params);
MOBILEGL_GL_API void glNamedProgramLocalParametersI4ivEXT(GLuint program, GLenum target, GLuint index, GLsizei count, const GLint* params);
MOBILEGL_GL_API void glNamedProgramLocalParametersI4uivEXT(GLuint program, GLenum target, GLuint index, GLsizei count, const GLuint* params);
MOBILEGL_GL_API void glNamedRenderbufferStorage(GLuint renderbuffer, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glNamedRenderbufferStorageEXT(GLuint renderbuffer, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glNamedRenderbufferStorageMultisample(GLuint renderbuffer, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glNamedRenderbufferStorageMultisampleAdvancedAMD(GLuint renderbuffer, GLsizei samples, GLsizei storageSamples, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glNamedRenderbufferStorageMultisampleCoverageEXT(GLuint renderbuffer, GLsizei coverageSamples, GLsizei colorSamples, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glNamedRenderbufferStorageMultisampleEXT(GLuint renderbuffer, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glNamedStringARB(GLenum type, GLint namelen, const GLchar* name, GLint strinen, const GLchar* string);
MOBILEGL_GL_API void glNormalFormatNV(GLenum type, GLsizei stride);
MOBILEGL_GL_API void glObjectLabel(GLenum identifier, GLuint name, GLsizei length, const GLchar* label);
MOBILEGL_GL_API void glObjectPtrLabel(const void* ptr, GLsizei length, const GLchar* label);
MOBILEGL_GL_API void glPatchParameteri(GLenum pname, GLint value);
MOBILEGL_GL_API void glPathCoverDepthFuncNV(GLenum func);
MOBILEGL_GL_API void glPathParameterfNV(GLuint path, GLenum pname, GLfloat value);
MOBILEGL_GL_API void glPathParameteriNV(GLuint path, GLenum pname, GLint value);
MOBILEGL_GL_API void glPathStencilDepthOffsetNV(GLfloat factor, GLfloat units);
MOBILEGL_GL_API void glPathStencilFuncNV(GLenum func, GLint ref, GLuint mask);
MOBILEGL_GL_API void glPathStringNV(GLuint path, GLenum format, GLsizei length, const void* pathString);
MOBILEGL_GL_API void glPixelStoref(GLenum pname, GLfloat param);
MOBILEGL_GL_API void glPixelStorei(GLenum pname, GLint param);
MOBILEGL_GL_API void glPointParameterf(GLenum pname, GLfloat param);
MOBILEGL_GL_API void glPointParameteri(GLenum pname, GLint param);
MOBILEGL_GL_API void glPointSize(GLfloat size);
MOBILEGL_GL_API void glPolygonMode(GLenum face, GLenum mode);
MOBILEGL_GL_API void glPolygonOffset(GLfloat factor, GLfloat units);
MOBILEGL_GL_API void glPolygonOffsetClamp(GLfloat factor, GLfloat units, GLfloat clamp);
MOBILEGL_GL_API void glPolygonOffsetClampEXT(GLfloat factor, GLfloat units, GLfloat clamp);
MOBILEGL_GL_API void glPrimitiveBoundingBoxARB(GLfloat minX, GLfloat minY, GLfloat minZ, GLfloat minW, GLfloat maxX, GLfloat maxY, GLfloat maxZ, GLfloat maxW);
MOBILEGL_GL_API void glPrimitiveRestartIndex(GLuint index);
MOBILEGL_GL_API void glProgramBinary(GLuint program, GLenum binaryFormat, const void* binary, GLsizei length);
MOBILEGL_GL_API void glProgramParameteri(GLuint program, GLenum pname, GLint value);
MOBILEGL_GL_API void glProgramUniform1d(GLuint program, GLint location, GLdouble v0);
MOBILEGL_GL_API void glProgramUniform1dEXT(GLuint program, GLint location, GLdouble x);
MOBILEGL_GL_API void glProgramUniform1dv(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform1dvEXT(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform1f(GLuint program, GLint location, GLfloat v0);
MOBILEGL_GL_API void glProgramUniform1fv(GLuint program, GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniform1i(GLuint program, GLint location, GLint v0);
MOBILEGL_GL_API void glProgramUniform1i64ARB(GLuint program, GLint location, GLint64 x);
MOBILEGL_GL_API void glProgramUniform1i64NV(GLuint program, GLint location, GLint64EXT x);
MOBILEGL_GL_API void glProgramUniform1i64vARB(GLuint program, GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glProgramUniform1i64vNV(GLuint program, GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glProgramUniform1iv(GLuint program, GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glProgramUniform1ui(GLuint program, GLint location, GLuint v0);
MOBILEGL_GL_API void glProgramUniform1ui64ARB(GLuint program, GLint location, GLuint64 x);
MOBILEGL_GL_API void glProgramUniform1ui64NV(GLuint program, GLint location, GLuint64EXT x);
MOBILEGL_GL_API void glProgramUniform1ui64vARB(GLuint program, GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glProgramUniform1ui64vNV(GLuint program, GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glProgramUniform1uiv(GLuint program, GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glProgramUniform2d(GLuint program, GLint location, GLdouble v0, GLdouble v1);
MOBILEGL_GL_API void glProgramUniform2dEXT(GLuint program, GLint location, GLdouble x, GLdouble y);
MOBILEGL_GL_API void glProgramUniform2dv(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform2dvEXT(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform2f(GLuint program, GLint location, GLfloat v0, GLfloat v1);
MOBILEGL_GL_API void glProgramUniform2fv(GLuint program, GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniform2i(GLuint program, GLint location, GLint v0, GLint v1);
MOBILEGL_GL_API void glProgramUniform2i64ARB(GLuint program, GLint location, GLint64 x, GLint64 y);
MOBILEGL_GL_API void glProgramUniform2i64NV(GLuint program, GLint location, GLint64EXT x, GLint64EXT y);
MOBILEGL_GL_API void glProgramUniform2i64vARB(GLuint program, GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glProgramUniform2i64vNV(GLuint program, GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glProgramUniform2iv(GLuint program, GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glProgramUniform2ui(GLuint program, GLint location, GLuint v0, GLuint v1);
MOBILEGL_GL_API void glProgramUniform2ui64ARB(GLuint program, GLint location, GLuint64 x, GLuint64 y);
MOBILEGL_GL_API void glProgramUniform2ui64NV(GLuint program, GLint location, GLuint64EXT x, GLuint64EXT y);
MOBILEGL_GL_API void glProgramUniform2ui64vARB(GLuint program, GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glProgramUniform2ui64vNV(GLuint program, GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glProgramUniform2uiv(GLuint program, GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glProgramUniform3d(GLuint program, GLint location, GLdouble v0, GLdouble v1, GLdouble v2);
MOBILEGL_GL_API void glProgramUniform3dEXT(GLuint program, GLint location, GLdouble x, GLdouble y, GLdouble z);
MOBILEGL_GL_API void glProgramUniform3dv(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform3dvEXT(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform3f(GLuint program, GLint location, GLfloat v0, GLfloat v1, GLfloat v2);
MOBILEGL_GL_API void glProgramUniform3fv(GLuint program, GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniform3i(GLuint program, GLint location, GLint v0, GLint v1, GLint v2);
MOBILEGL_GL_API void glProgramUniform3i64ARB(GLuint program, GLint location, GLint64 x, GLint64 y, GLint64 z);
MOBILEGL_GL_API void glProgramUniform3i64NV(GLuint program, GLint location, GLint64EXT x, GLint64EXT y, GLint64EXT z);
MOBILEGL_GL_API void glProgramUniform3i64vARB(GLuint program, GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glProgramUniform3i64vNV(GLuint program, GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glProgramUniform3iv(GLuint program, GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glProgramUniform3ui(GLuint program, GLint location, GLuint v0, GLuint v1, GLuint v2);
MOBILEGL_GL_API void glProgramUniform3ui64ARB(GLuint program, GLint location, GLuint64 x, GLuint64 y, GLuint64 z);
MOBILEGL_GL_API void glProgramUniform3ui64NV(GLuint program, GLint location, GLuint64EXT x, GLuint64EXT y, GLuint64EXT z);
MOBILEGL_GL_API void glProgramUniform3ui64vARB(GLuint program, GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glProgramUniform3ui64vNV(GLuint program, GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glProgramUniform3uiv(GLuint program, GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glProgramUniform4d(GLuint program, GLint location, GLdouble v0, GLdouble v1, GLdouble v2, GLdouble v3);
MOBILEGL_GL_API void glProgramUniform4dEXT(GLuint program, GLint location, GLdouble x, GLdouble y, GLdouble z, GLdouble w);
MOBILEGL_GL_API void glProgramUniform4dv(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform4dvEXT(GLuint program, GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniform4f(GLuint program, GLint location, GLfloat v0, GLfloat v1, GLfloat v2, GLfloat v3);
MOBILEGL_GL_API void glProgramUniform4fv(GLuint program, GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniform4i(GLuint program, GLint location, GLint v0, GLint v1, GLint v2, GLint v3);
MOBILEGL_GL_API void glProgramUniform4i64ARB(GLuint program, GLint location, GLint64 x, GLint64 y, GLint64 z, GLint64 w);
MOBILEGL_GL_API void glProgramUniform4i64NV(GLuint program, GLint location, GLint64EXT x, GLint64EXT y, GLint64EXT z, GLint64EXT w);
MOBILEGL_GL_API void glProgramUniform4i64vARB(GLuint program, GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glProgramUniform4i64vNV(GLuint program, GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glProgramUniform4iv(GLuint program, GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glProgramUniform4ui(GLuint program, GLint location, GLuint v0, GLuint v1, GLuint v2, GLuint v3);
MOBILEGL_GL_API void glProgramUniform4ui64ARB(GLuint program, GLint location, GLuint64 x, GLuint64 y, GLuint64 z, GLuint64 w);
MOBILEGL_GL_API void glProgramUniform4ui64NV(GLuint program, GLint location, GLuint64EXT x, GLuint64EXT y, GLuint64EXT z, GLuint64EXT w);
MOBILEGL_GL_API void glProgramUniform4ui64vARB(GLuint program, GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glProgramUniform4ui64vNV(GLuint program, GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glProgramUniform4uiv(GLuint program, GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glProgramUniformHandleui64ARB(GLuint program, GLint location, GLuint64 value);
MOBILEGL_GL_API void glProgramUniformHandleui64NV(GLuint program, GLint location, GLuint64 value);
MOBILEGL_GL_API void glProgramUniformHandleui64vARB(GLuint program, GLint location, GLsizei count, const GLuint64* values);
MOBILEGL_GL_API void glProgramUniformHandleui64vNV(GLuint program, GLint location, GLsizei count, const GLuint64* values);
MOBILEGL_GL_API void glProgramUniformMatrix2dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix2dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix2fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix2x3dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix2x3dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix2x3fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix2x4dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix2x4dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix2x4fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix3dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix3dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix3fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix3x2dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix3x2dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix3x2fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix3x4dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix3x4dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix3x4fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix4dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix4dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix4fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix4x2dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix4x2dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix4x2fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformMatrix4x3dv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix4x3dvEXT(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glProgramUniformMatrix4x3fv(GLuint program, GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glProgramUniformui64NV(GLuint program, GLint location, GLuint64EXT value);
MOBILEGL_GL_API void glProgramUniformui64vNV(GLuint program, GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glProvokingVertex(GLenum mode);
MOBILEGL_GL_API void glPushClientAttribDefaultEXT(GLbitfield mask);
MOBILEGL_GL_API void glPushDebugGroup(GLenum source, GLuint id, GLsizei length, const GLchar* message);
MOBILEGL_GL_API void glPushGroupMarkerEXT(GLsizei length, const GLchar* marker);
MOBILEGL_GL_API void glQueryCounter(GLuint id, GLenum target);
MOBILEGL_GL_API void glRasterSamplesEXT(GLuint samples, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glReadBuffer(GLenum src);
MOBILEGL_GL_API void glRenderbufferStorage(GLenum target, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glRenderbufferStorageMultisample(GLenum target, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glRenderbufferStorageMultisampleAdvancedAMD(GLenum target, GLsizei samples, GLsizei storageSamples, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glRenderbufferStorageMultisampleCoverageNV(GLenum target, GLsizei coverageSamples, GLsizei colorSamples, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glResetMemoryObjectParameterNV(GLuint memory, GLenum pname);
MOBILEGL_GL_API void glSampleCoverage(GLfloat value, GLboolean invert);
MOBILEGL_GL_API void glSampleMaski(GLuint maskNumber, GLbitfield mask);
MOBILEGL_GL_API void glSamplerParameterf(GLuint sampler, GLenum pname, GLfloat param);
MOBILEGL_GL_API void glSamplerParameteri(GLuint sampler, GLenum pname, GLint param);
MOBILEGL_GL_API void glScissor(GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glScissorArrayv(GLuint first, GLsizei count, const GLint* v);
MOBILEGL_GL_API void glScissorExclusiveArrayvNV(GLuint first, GLsizei count, const GLint* v);
MOBILEGL_GL_API void glScissorExclusiveNV(GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glScissorIndexed(GLuint index, GLint left, GLint bottom, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glSecondaryColorFormatNV(GLint size, GLenum type, GLsizei stride);
MOBILEGL_GL_API void glShaderBinary(GLsizei count, const GLuint* shaders, GLenum binaryformat, const void* binary, GLsizei length);
MOBILEGL_GL_API void glShaderStorageBlockBinding(GLuint program, GLuint storageBlockIndex, GLuint storageBlockBinding);
MOBILEGL_GL_API void glShadingRateImageBarrierNV(GLboolean synchronize);
MOBILEGL_GL_API void glShadingRateImagePaletteNV(GLuint viewport, GLuint first, GLsizei count, const GLenum* rates);
MOBILEGL_GL_API void glShadingRateSampleOrderNV(GLenum order);
MOBILEGL_GL_API void glSignalVkFenceNV(GLuint64 vkFence);
MOBILEGL_GL_API void glSignalVkSemaphoreNV(GLuint64 vkSemaphore);
MOBILEGL_GL_API void glStateCaptureNV(GLuint state, GLenum mode);
MOBILEGL_GL_API void glStencilFillPathNV(GLuint path, GLenum fillMode, GLuint mask);
MOBILEGL_GL_API void glStencilFunc(GLenum func, GLint ref, GLuint mask);
MOBILEGL_GL_API void glStencilFuncSeparate(GLenum face, GLenum func, GLint ref, GLuint mask);
MOBILEGL_GL_API void glStencilMask(GLuint mask);
MOBILEGL_GL_API void glStencilMaskSeparate(GLenum face, GLuint mask);
MOBILEGL_GL_API void glStencilOp(GLenum fail, GLenum zfail, GLenum zpass);
MOBILEGL_GL_API void glStencilOpSeparate(GLenum face, GLenum sfail, GLenum dpfail, GLenum dppass);
MOBILEGL_GL_API void glStencilStrokePathNV(GLuint path, GLint reference, GLuint mask);
MOBILEGL_GL_API void glStencilThenCoverFillPathNV(GLuint path, GLenum fillMode, GLuint mask, GLenum coverMode);
MOBILEGL_GL_API void glStencilThenCoverStrokePathNV(GLuint path, GLint reference, GLuint mask, GLenum coverMode);
MOBILEGL_GL_API void glSubpixelPrecisionBiasNV(GLuint xbits, GLuint ybits);
MOBILEGL_GL_API void glTexAttachMemoryNV(GLenum target, GLuint memory, GLuint64 offset);
MOBILEGL_GL_API void glTexBuffer(GLenum target, GLenum internalformat, GLuint buffer);
MOBILEGL_GL_API void glTexBufferRange(GLenum target, GLenum internalformat, GLuint buffer, GLintptr offset, GLsizeiptr size);
MOBILEGL_GL_API void glTexCoordFormatNV(GLint size, GLenum type, GLsizei stride);
MOBILEGL_GL_API void glTexImage2DMultisample(GLenum target, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTexImage3DMultisample(GLenum target, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTexPageCommitmentARB(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLboolean commit);
MOBILEGL_GL_API void glTexPageCommitmentMemNV(GLenum target, GLint layer, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLuint memory, GLuint64 offset, GLboolean commit);
MOBILEGL_GL_API void glTexParameterf(GLenum target, GLenum pname, GLfloat param);
MOBILEGL_GL_API void glTexParameteri(GLenum target, GLenum pname, GLint param);
MOBILEGL_GL_API void glTexStorage1DEXT(GLenum target, GLsizei levels, GLenum internalformat, GLsizei width);
MOBILEGL_GL_API void glTexStorage2D(GLenum target, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glTexStorage2DMultisample(GLenum target, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTexStorage3D(GLenum target, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth);
MOBILEGL_GL_API void glTexStorage3DMultisample(GLenum target, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTextureAttachMemoryNV(GLuint texture, GLuint memory, GLuint64 offset);
MOBILEGL_GL_API void glTextureBuffer(GLuint texture, GLenum internalformat, GLuint buffer);
MOBILEGL_GL_API void glTextureBufferEXT(GLuint texture, GLenum target, GLenum internalformat, GLuint buffer);
MOBILEGL_GL_API void glTextureBufferRange(GLuint texture, GLenum internalformat, GLuint buffer, GLintptr offset, GLsizeiptr size);
MOBILEGL_GL_API void glTextureBufferRangeEXT(GLuint texture, GLenum target, GLenum internalformat, GLuint buffer, GLintptr offset, GLsizeiptr size);
MOBILEGL_GL_API void glTexturePageCommitmentEXT(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLboolean commit);
MOBILEGL_GL_API void glTexturePageCommitmentMemNV(GLuint texture, GLint layer, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLuint memory, GLuint64 offset, GLboolean commit);
MOBILEGL_GL_API void glTextureParameterf(GLuint texture, GLenum pname, GLfloat param);
MOBILEGL_GL_API void glTextureParameterfEXT(GLuint texture, GLenum target, GLenum pname, GLfloat param);
MOBILEGL_GL_API void glTextureParameteri(GLuint texture, GLenum pname, GLint param);
MOBILEGL_GL_API void glTextureParameteriEXT(GLuint texture, GLenum target, GLenum pname, GLint param);
MOBILEGL_GL_API void glTextureRenderbufferEXT(GLuint texture, GLenum target, GLuint renderbuffer);
MOBILEGL_GL_API void glTextureStorage1D(GLuint texture, GLsizei levels, GLenum internalformat, GLsizei width);
MOBILEGL_GL_API void glTextureStorage1DEXT(GLuint texture, GLenum target, GLsizei levels, GLenum internalformat, GLsizei width);
MOBILEGL_GL_API void glTextureStorage2D(GLuint texture, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glTextureStorage2DEXT(GLuint texture, GLenum target, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glTextureStorage2DMultisample(GLuint texture, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTextureStorage2DMultisampleEXT(GLuint texture, GLenum target, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTextureStorage3D(GLuint texture, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth);
MOBILEGL_GL_API void glTextureStorage3DEXT(GLuint texture, GLenum target, GLsizei levels, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth);
MOBILEGL_GL_API void glTextureStorage3DMultisample(GLuint texture, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTextureStorage3DMultisampleEXT(GLuint texture, GLenum target, GLsizei samples, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLboolean fixedsamplelocations);
MOBILEGL_GL_API void glTextureView(GLuint texture, GLenum target, GLuint origtexture, GLenum internalformat, GLuint minlevel, GLuint numlevels, GLuint minlayer, GLuint numlayers);
MOBILEGL_GL_API void glTransformFeedbackBufferBase(GLuint xfb, GLuint index, GLuint buffer);
MOBILEGL_GL_API void glTransformFeedbackBufferRange(GLuint xfb, GLuint index, GLuint buffer, GLintptr offset, GLsizeiptr size);
MOBILEGL_GL_API void glUniform1d(GLint location, GLdouble x);
MOBILEGL_GL_API void glUniform1dv(GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glUniform1f(GLint location, GLfloat v0);
MOBILEGL_GL_API void glUniform1fv(GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glUniform1i(GLint location, GLint v0);
MOBILEGL_GL_API void glUniform1i64ARB(GLint location, GLint64 x);
MOBILEGL_GL_API void glUniform1i64NV(GLint location, GLint64EXT x);
MOBILEGL_GL_API void glUniform1i64vARB(GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glUniform1i64vNV(GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glUniform1iv(GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glUniform1ui(GLint location, GLuint v0);
MOBILEGL_GL_API void glUniform1ui64ARB(GLint location, GLuint64 x);
MOBILEGL_GL_API void glUniform1ui64NV(GLint location, GLuint64EXT x);
MOBILEGL_GL_API void glUniform1ui64vARB(GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glUniform1ui64vNV(GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glUniform1uiv(GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glUniform2d(GLint location, GLdouble x, GLdouble y);
MOBILEGL_GL_API void glUniform2dv(GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glUniform2f(GLint location, GLfloat v0, GLfloat v1);
MOBILEGL_GL_API void glUniform2fv(GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glUniform2i(GLint location, GLint v0, GLint v1);
MOBILEGL_GL_API void glUniform2i64ARB(GLint location, GLint64 x, GLint64 y);
MOBILEGL_GL_API void glUniform2i64NV(GLint location, GLint64EXT x, GLint64EXT y);
MOBILEGL_GL_API void glUniform2i64vARB(GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glUniform2i64vNV(GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glUniform2iv(GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glUniform2ui(GLint location, GLuint v0, GLuint v1);
MOBILEGL_GL_API void glUniform2ui64ARB(GLint location, GLuint64 x, GLuint64 y);
MOBILEGL_GL_API void glUniform2ui64NV(GLint location, GLuint64EXT x, GLuint64EXT y);
MOBILEGL_GL_API void glUniform2ui64vARB(GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glUniform2ui64vNV(GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glUniform2uiv(GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glUniform3d(GLint location, GLdouble x, GLdouble y, GLdouble z);
MOBILEGL_GL_API void glUniform3dv(GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glUniform3f(GLint location, GLfloat v0, GLfloat v1, GLfloat v2);
MOBILEGL_GL_API void glUniform3fv(GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glUniform3i(GLint location, GLint v0, GLint v1, GLint v2);
MOBILEGL_GL_API void glUniform3i64ARB(GLint location, GLint64 x, GLint64 y, GLint64 z);
MOBILEGL_GL_API void glUniform3i64NV(GLint location, GLint64EXT x, GLint64EXT y, GLint64EXT z);
MOBILEGL_GL_API void glUniform3i64vARB(GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glUniform3i64vNV(GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glUniform3iv(GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glUniform3ui(GLint location, GLuint v0, GLuint v1, GLuint v2);
MOBILEGL_GL_API void glUniform3ui64ARB(GLint location, GLuint64 x, GLuint64 y, GLuint64 z);
MOBILEGL_GL_API void glUniform3ui64NV(GLint location, GLuint64EXT x, GLuint64EXT y, GLuint64EXT z);
MOBILEGL_GL_API void glUniform3ui64vARB(GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glUniform3ui64vNV(GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glUniform3uiv(GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glUniform4d(GLint location, GLdouble x, GLdouble y, GLdouble z, GLdouble w);
MOBILEGL_GL_API void glUniform4dv(GLint location, GLsizei count, const GLdouble* value);
MOBILEGL_GL_API void glUniform4f(GLint location, GLfloat v0, GLfloat v1, GLfloat v2, GLfloat v3);
MOBILEGL_GL_API void glUniform4fv(GLint location, GLsizei count, const GLfloat* value);
MOBILEGL_GL_API void glUniform4i(GLint location, GLint v0, GLint v1, GLint v2, GLint v3);
MOBILEGL_GL_API void glUniform4i64ARB(GLint location, GLint64 x, GLint64 y, GLint64 z, GLint64 w);
MOBILEGL_GL_API void glUniform4i64NV(GLint location, GLint64EXT x, GLint64EXT y, GLint64EXT z, GLint64EXT w);
MOBILEGL_GL_API void glUniform4i64vARB(GLint location, GLsizei count, const GLint64* value);
MOBILEGL_GL_API void glUniform4i64vNV(GLint location, GLsizei count, const GLint64EXT* value);
MOBILEGL_GL_API void glUniform4iv(GLint location, GLsizei count, const GLint* value);
MOBILEGL_GL_API void glUniform4ui(GLint location, GLuint v0, GLuint v1, GLuint v2, GLuint v3);
MOBILEGL_GL_API void glUniform4ui64ARB(GLint location, GLuint64 x, GLuint64 y, GLuint64 z, GLuint64 w);
MOBILEGL_GL_API void glUniform4ui64NV(GLint location, GLuint64EXT x, GLuint64EXT y, GLuint64EXT z, GLuint64EXT w);
MOBILEGL_GL_API void glUniform4ui64vARB(GLint location, GLsizei count, const GLuint64* value);
MOBILEGL_GL_API void glUniform4ui64vNV(GLint location, GLsizei count, const GLuint64EXT* value);
MOBILEGL_GL_API void glUniform4uiv(GLint location, GLsizei count, const GLuint* value);
MOBILEGL_GL_API void glUniformBlockBinding(GLuint program, GLuint uniformBlockIndex, GLuint uniformBlockBinding);
MOBILEGL_GL_API void glUniformHandleui64ARB(GLint location, GLuint64 value);
MOBILEGL_GL_API void glUniformHandleui64NV(GLint location, GLuint64 value);
MOBILEGL_GL_API void glUniformMatrix2dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix2fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix2x3dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix2x3fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix2x4dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix2x4fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix3dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix3fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix3x2dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix3x2fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix3x4dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix3x4fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix4dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix4fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix4x2dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix4x2fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformMatrix4x3dv(GLint location, GLsizei count, GLboolean transpose, const GLdouble* value);
MOBILEGL_GL_API void glUniformMatrix4x3fv(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value);
MOBILEGL_GL_API void glUniformui64NV(GLint location, GLuint64EXT value);
MOBILEGL_GL_API GLboolean glUnmapBuffer(GLenum target);
MOBILEGL_GL_API GLboolean glUnmapNamedBuffer(GLuint buffer);
MOBILEGL_GL_API GLboolean glUnmapNamedBufferEXT(GLuint buffer);
MOBILEGL_GL_API void glUseProgram(GLuint program);
MOBILEGL_GL_API void glUseProgramStages(GLuint pipeline, GLbitfield stages, GLuint program);
MOBILEGL_GL_API void glUseShaderProgramEXT(GLenum type, GLuint program);
MOBILEGL_GL_API void glValidateProgram(GLuint program);
MOBILEGL_GL_API void glValidateProgramPipeline(GLuint pipeline);
MOBILEGL_GL_API void glVertexArrayAttribBinding(GLuint vaobj, GLuint attribindex, GLuint bindingindex);
MOBILEGL_GL_API void glVertexArrayAttribFormat(GLuint vaobj, GLuint attribindex, GLint size, GLenum type, GLboolean normalized, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexArrayAttribIFormat(GLuint vaobj, GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexArrayAttribLFormat(GLuint vaobj, GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexArrayBindVertexBufferEXT(GLuint vaobj, GLuint bindingindex, GLuint buffer, GLintptr offset, GLsizei stride);
MOBILEGL_GL_API void glVertexArrayBindingDivisor(GLuint vaobj, GLuint bindingindex, GLuint divisor);
MOBILEGL_GL_API void glVertexArrayColorOffsetEXT(GLuint vaobj, GLuint buffer, GLint size, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayEdgeFlagOffsetEXT(GLuint vaobj, GLuint buffer, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayElementBuffer(GLuint vaobj, GLuint buffer);
MOBILEGL_GL_API void glVertexArrayFogCoordOffsetEXT(GLuint vaobj, GLuint buffer, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayIndexOffsetEXT(GLuint vaobj, GLuint buffer, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayMultiTexCoordOffsetEXT(GLuint vaobj, GLuint buffer, GLenum texunit, GLint size, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayNormalOffsetEXT(GLuint vaobj, GLuint buffer, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArraySecondaryColorOffsetEXT(GLuint vaobj, GLuint buffer, GLint size, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayTexCoordOffsetEXT(GLuint vaobj, GLuint buffer, GLint size, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayVertexAttribBindingEXT(GLuint vaobj, GLuint attribindex, GLuint bindingindex);
MOBILEGL_GL_API void glVertexArrayVertexAttribDivisorEXT(GLuint vaobj, GLuint index, GLuint divisor);
MOBILEGL_GL_API void glVertexArrayVertexAttribFormatEXT(GLuint vaobj, GLuint attribindex, GLint size, GLenum type, GLboolean normalized, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexArrayVertexAttribIFormatEXT(GLuint vaobj, GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexArrayVertexAttribIOffsetEXT(GLuint vaobj, GLuint buffer, GLuint index, GLint size, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayVertexAttribLFormatEXT(GLuint vaobj, GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexArrayVertexAttribLOffsetEXT(GLuint vaobj, GLuint buffer, GLuint index, GLint size, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayVertexAttribOffsetEXT(GLuint vaobj, GLuint buffer, GLuint index, GLint size, GLenum type, GLboolean normalized, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexArrayVertexBindingDivisorEXT(GLuint vaobj, GLuint bindingindex, GLuint divisor);
MOBILEGL_GL_API void glVertexArrayVertexBuffer(GLuint vaobj, GLuint bindingindex, GLuint buffer, GLintptr offset, GLsizei stride);
MOBILEGL_GL_API void glVertexArrayVertexOffsetEXT(GLuint vaobj, GLuint buffer, GLint size, GLenum type, GLsizei stride, GLintptr offset);
MOBILEGL_GL_API void glVertexAttrib1d(GLuint index, GLdouble x);
MOBILEGL_GL_API void glVertexAttrib1f(GLuint index, GLfloat x);
MOBILEGL_GL_API void glVertexAttrib1s(GLuint index, GLshort x);
MOBILEGL_GL_API void glVertexAttrib2d(GLuint index, GLdouble x, GLdouble y);
MOBILEGL_GL_API void glVertexAttrib2f(GLuint index, GLfloat x, GLfloat y);
MOBILEGL_GL_API void glVertexAttrib2s(GLuint index, GLshort x, GLshort y);
MOBILEGL_GL_API void glVertexAttrib3d(GLuint index, GLdouble x, GLdouble y, GLdouble z);
MOBILEGL_GL_API void glVertexAttrib3f(GLuint index, GLfloat x, GLfloat y, GLfloat z);
MOBILEGL_GL_API void glVertexAttrib3s(GLuint index, GLshort x, GLshort y, GLshort z);
MOBILEGL_GL_API void glVertexAttrib4Nub(GLuint index, GLubyte x, GLubyte y, GLubyte z, GLubyte w);
MOBILEGL_GL_API void glVertexAttrib4d(GLuint index, GLdouble x, GLdouble y, GLdouble z, GLdouble w);
MOBILEGL_GL_API void glVertexAttrib4f(GLuint index, GLfloat x, GLfloat y, GLfloat z, GLfloat w);
MOBILEGL_GL_API void glVertexAttrib4s(GLuint index, GLshort x, GLshort y, GLshort z, GLshort w);
MOBILEGL_GL_API void glVertexAttribBinding(GLuint attribindex, GLuint bindingindex);
MOBILEGL_GL_API void glVertexAttribDivisor(GLuint index, GLuint divisor);
MOBILEGL_GL_API void glVertexAttribFormat(GLuint attribindex, GLint size, GLenum type, GLboolean normalized, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexAttribFormatNV(GLuint index, GLint size, GLenum type, GLboolean normalized, GLsizei stride);
MOBILEGL_GL_API void glVertexAttribI1i(GLuint index, GLint x);
MOBILEGL_GL_API void glVertexAttribI1ui(GLuint index, GLuint x);
MOBILEGL_GL_API void glVertexAttribI2i(GLuint index, GLint x, GLint y);
MOBILEGL_GL_API void glVertexAttribI2ui(GLuint index, GLuint x, GLuint y);
MOBILEGL_GL_API void glVertexAttribI3i(GLuint index, GLint x, GLint y, GLint z);
MOBILEGL_GL_API void glVertexAttribI3ui(GLuint index, GLuint x, GLuint y, GLuint z);
MOBILEGL_GL_API void glVertexAttribI4i(GLuint index, GLint x, GLint y, GLint z, GLint w);
MOBILEGL_GL_API void glVertexAttribI4ui(GLuint index, GLuint x, GLuint y, GLuint z, GLuint w);
MOBILEGL_GL_API void glVertexAttribIFormat(GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexAttribIFormatNV(GLuint index, GLint size, GLenum type, GLsizei stride);
MOBILEGL_GL_API void glVertexAttribIPointer(GLuint index, GLint size, GLenum type, GLsizei stride, const void* pointer);
MOBILEGL_GL_API void glVertexAttribL1d(GLuint index, GLdouble x);
MOBILEGL_GL_API void glVertexAttribL1i64NV(GLuint index, GLint64EXT x);
MOBILEGL_GL_API void glVertexAttribL1ui64ARB(GLuint index, GLuint64EXT x);
MOBILEGL_GL_API void glVertexAttribL1ui64NV(GLuint index, GLuint64EXT x);
MOBILEGL_GL_API void glVertexAttribL2d(GLuint index, GLdouble x, GLdouble y);
MOBILEGL_GL_API void glVertexAttribL2i64NV(GLuint index, GLint64EXT x, GLint64EXT y);
MOBILEGL_GL_API void glVertexAttribL2ui64NV(GLuint index, GLuint64EXT x, GLuint64EXT y);
MOBILEGL_GL_API void glVertexAttribL3d(GLuint index, GLdouble x, GLdouble y, GLdouble z);
MOBILEGL_GL_API void glVertexAttribL3i64NV(GLuint index, GLint64EXT x, GLint64EXT y, GLint64EXT z);
MOBILEGL_GL_API void glVertexAttribL3ui64NV(GLuint index, GLuint64EXT x, GLuint64EXT y, GLuint64EXT z);
MOBILEGL_GL_API void glVertexAttribL4d(GLuint index, GLdouble x, GLdouble y, GLdouble z, GLdouble w);
MOBILEGL_GL_API void glVertexAttribL4i64NV(GLuint index, GLint64EXT x, GLint64EXT y, GLint64EXT z, GLint64EXT w);
MOBILEGL_GL_API void glVertexAttribL4ui64NV(GLuint index, GLuint64EXT x, GLuint64EXT y, GLuint64EXT z, GLuint64EXT w);
MOBILEGL_GL_API void glVertexAttribLFormat(GLuint attribindex, GLint size, GLenum type, GLuint relativeoffset);
MOBILEGL_GL_API void glVertexAttribLFormatNV(GLuint index, GLint size, GLenum type, GLsizei stride);
MOBILEGL_GL_API void glVertexAttribLPointer(GLuint index, GLint size, GLenum type, GLsizei stride, const void* pointer);
MOBILEGL_GL_API void glVertexAttribP1ui(GLuint index, GLenum type, GLboolean normalized, GLuint value);
MOBILEGL_GL_API void glVertexAttribP2ui(GLuint index, GLenum type, GLboolean normalized, GLuint value);
MOBILEGL_GL_API void glVertexAttribP3ui(GLuint index, GLenum type, GLboolean normalized, GLuint value);
MOBILEGL_GL_API void glVertexAttribP4ui(GLuint index, GLenum type, GLboolean normalized, GLuint value);
MOBILEGL_GL_API void glVertexAttribPointer(GLuint index, GLint size, GLenum type, GLboolean normalized, GLsizei stride, const void* pointer);
MOBILEGL_GL_API void glVertexBindingDivisor(GLuint bindingindex, GLuint divisor);
MOBILEGL_GL_API void glVertexFormatNV(GLint size, GLenum type, GLsizei stride);
MOBILEGL_GL_API void glViewport(GLint x, GLint y, GLsizei width, GLsizei height);
MOBILEGL_GL_API void glViewportArrayv(GLuint first, GLsizei count, const GLfloat* v);
MOBILEGL_GL_API void glViewportIndexedf(GLuint index, GLfloat x, GLfloat y, GLfloat w, GLfloat h);
MOBILEGL_GL_API void glViewportPositionWScaleNV(GLuint index, GLfloat xcoeff, GLfloat ycoeff);
MOBILEGL_GL_API void glViewportSwizzleNV(GLuint index, GLenum swizzlex, GLenum swizzley, GLenum swizzlez, GLenum swizzlew);
MOBILEGL_GL_API void glWaitSync(GLsync sync, GLbitfield flags, GLuint64 timeout);
MOBILEGL_GL_API void glWaitVkSemaphoreNV(GLuint64 vkSemaphore);
namespace MobileGL::ClientTrampoline::Gen {
// Resolves a GL function name against the exported trampoline table
// (direct &glXxx references - no dlsym, mirroring monolith
// MobileGL::MG_Impl::GetProcAddress semantics).
void* LookupGeneratedProc(const char* name);
}
// End of File
+663
View File
@@ -0,0 +1,663 @@
// MobileGL - MobileGL/MG_Gbm/GbmApi.cpp
// SPDX-License-Identifier: LGPL-3.0-only
//
// The extern "C" surface: every entry point gbm.h declares, and nothing else.
//
// This file is a translation layer, and it is kept that way on purpose. Each entry
// point does three things - check the pointers it was handed, take the device's one
// mutex for the whole call, and hand the work to GbmDevice.cpp - so the concurrency
// rule ("one mutex per device, taken once at the API boundary, never inside the
// allocation code") is visible in one place instead of being spread over the
// backends. A missing lock is a bug that only shows up in the compositor, so the
// boilerplate is deliberately identical in every function.
//
// Two conventions are observable from outside and are stated here because they are
// the ones a caller can get wrong:
//
// * gbm_bo_get_fd and gbm_bo_get_fd_for_plane return a DUP the caller owns. The
// caller closes what it was given; the buffer's own descriptor is closed once,
// when the buffer is destroyed, and never by a caller's close.
// * Nothing here throws and nothing here aborts. A failure is reported the way
// GBM reports failures: null or -1 with errno set, so that the strerror(errno)
// a caller prints describes the failure that actually happened.
#include "GbmInternal.h"
#include <cerrno>
using namespace MobileGL;
using namespace MobileGL::MG_Gbm;
namespace
{
// The one mutex that guards a record, reached through the device that owns it.
// Null means the pointer was not one this library handed out, and every entry
// point turns that into its own failure answer rather than dereferencing it.
//
// A buffer or surface with no owner cannot exist while the library is used
// correctly - the device destroys everything it still holds before it goes away
// - so null here is a defence against a stale pointer rather than a state the
// code produces.
std::mutex* MutexOf(struct gbm_device* device) noexcept
{
return device != nullptr ? &device->Impl.Mutex : nullptr;
}
std::mutex* MutexOf(struct gbm_bo* bo) noexcept
{
return bo != nullptr ? MutexOf(bo->Impl.Owner) : nullptr;
}
std::mutex* MutexOf(struct gbm_surface* surface) noexcept
{
return surface != nullptr ? MutexOf(surface->Impl.Owner) : nullptr;
}
} // namespace
// ---------------------------------------------------------------------------
// Device
// ---------------------------------------------------------------------------
extern "C" struct gbm_device* gbm_create_device(int fd)
{
// No lock to take: the device being built is not reachable by any other thread
// until this call returns it.
return GbmDeviceCreate(fd);
}
extern "C" void gbm_device_destroy(struct gbm_device* gbm)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
return;
}
// The lock is taken and deliberately never released: the mutex lives inside the
// record this call deletes, and unlocking a destroyed mutex is worse than
// leaving a dead one locked. Any other thread touching this device after its
// owner declared it dead is already outside what GBM can be asked to survive.
mutex->lock();
GbmDeviceDestroy(*gbm);
}
extern "C" int gbm_device_get_fd(struct gbm_device* gbm)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return -1;
}
std::lock_guard<std::mutex> guard(*mutex);
// The device's own dup of the fd it was created with, or -1 when it was created
// without one and no render node could be opened. That is not an error state:
// the self-allocated backend never talks to the node, so a device can be fully
// usable and still have no descriptor to report.
return gbm->Impl.DeviceFd;
}
extern "C" const char* gbm_device_get_backend_name(struct gbm_device* gbm)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
// The name says which memory this device hands out ("mobilegl-dma-heap" or
// "mobilegl-memfd"), because the callers that read it - a compositor logging its
// backend, a peer test recording what it got - are asking exactly that. The
// string lives in the device and outlives every call that can read it.
return gbm->Impl.BackendName.c_str();
}
extern "C" int gbm_device_is_format_supported(struct gbm_device* gbm, uint32_t format, uint32_t flags)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
// Every buffer this device produces is linear, mappable and shareable, so the
// usage flags do not narrow the answer to any of the usual combinations -
// including GBM_BO_USE_SCANOUT, which a linear dma-buf from the system heap
// satisfies. The one request the backend cannot honour is protected memory.
if ((flags & GBM_BO_USE_PROTECTED) != 0)
{
return 0;
}
return GbmFormatIsSupported(format) ? 1 : 0;
}
extern "C" int gbm_device_get_format_modifier_plane_count(struct gbm_device* gbm, uint32_t format, uint64_t modifier)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
// Only linear memory exists here, so any other modifier names a buffer this
// device will never produce and the answer is zero planes. INVALID is answered
// as well, because it means "no modifier was declared", which is the same
// layout question asked without an answer.
if (modifier != DrmFormatModLinear && modifier != DrmFormatModInvalid)
{
return 0;
}
GbmPlaneLayout planes[GbmMaxPlanes] = {};
Int planeCount = 0;
SizeT totalBytes = 0;
if (!GbmDescribeFormat(format, 1, 1, planes, planeCount, totalBytes))
{
return 0;
}
return planeCount;
}
// ---------------------------------------------------------------------------
// Buffer objects
// ---------------------------------------------------------------------------
extern "C" struct gbm_bo* gbm_bo_create(struct gbm_device* gbm, uint32_t width, uint32_t height, uint32_t format,
uint32_t flags)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
// No modifier list: the buffer is linear and gbm_bo_get_modifier says so.
return GbmBufferCreate(*gbm, width, height, format, flags, nullptr, 0);
}
extern "C" struct gbm_bo* gbm_bo_create_with_modifiers(struct gbm_device* gbm, uint32_t width, uint32_t height,
uint32_t format, const uint64_t* modifiers,
const unsigned int count)
{
// The modifier-less variant is the one with no flags parameter, so it is the
// flags-0 case of the other, and saying that here keeps one implementation.
return gbm_bo_create_with_modifiers2(gbm, width, height, format, modifiers, count, 0);
}
extern "C" struct gbm_bo* gbm_bo_create_with_modifiers2(struct gbm_device* gbm, uint32_t width, uint32_t height,
uint32_t format, const uint64_t* modifiers,
const unsigned int count, uint32_t flags)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
return GbmBufferCreate(*gbm, width, height, format, flags, modifiers, (Uint32)count);
}
extern "C" struct gbm_bo* gbm_bo_import(struct gbm_device* gbm, uint32_t type, void* buffer, uint32_t flags)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
return GbmBufferImport(*gbm, type, buffer, flags);
}
extern "C" void* gbm_bo_map(struct gbm_bo* bo, uint32_t x, uint32_t y, uint32_t width, uint32_t height,
uint32_t flags, uint32_t* stride, void** map_data)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
// The rectangle's size is not used to size the mapping: the whole buffer is
// mapped once, at its own length, and the rectangle is applied as an offset into
// it. That is what lets two overlapping maps share one address, and it is why a
// buffer's mapping stays valid for a caller that only ever asks for a corner.
(void)width;
(void)height;
(void)flags;
std::lock_guard<std::mutex> guard(*mutex);
Uint32 mapStride = 0;
void* address = GbmBufferMap(*bo, x, y, mapStride);
if (address == nullptr)
{
return nullptr;
}
if (stride != nullptr)
{
*stride = mapStride;
}
if (map_data != nullptr)
{
// The same address the function returns. Callers pass back either one to
// gbm_bo_unmap, and both have to name this mapping.
*map_data = address;
}
return address;
}
extern "C" void gbm_bo_unmap(struct gbm_bo* bo, void* map_data)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
return;
}
std::lock_guard<std::mutex> guard(*mutex);
GbmBufferUnmap(*bo, map_data);
}
extern "C" uint32_t gbm_bo_get_width(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
return bo->Impl.Width;
}
extern "C" uint32_t gbm_bo_get_height(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
return bo->Impl.Height;
}
extern "C" uint32_t gbm_bo_get_stride(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
return bo->Impl.Strides[0];
}
extern "C" uint32_t gbm_bo_get_stride_for_plane(struct gbm_bo* bo, int plane)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
// A plane this buffer does not have is answered with zero, which is the only
// value a caller can tell apart from a real stride.
if (plane < 0 || plane >= bo->Impl.PlaneCount)
{
errno = EINVAL;
return 0;
}
return bo->Impl.Strides[plane];
}
extern "C" uint32_t gbm_bo_get_format(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
return bo->Impl.Format;
}
extern "C" uint32_t gbm_bo_get_bpp(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
// Bits per pixel: a 32 bit format answers 32, which is what a caller computing a
// row length in bits or a row length in bytes from the same number expects.
return bo->Impl.BytesPerPixel * 8;
}
extern "C" uint32_t gbm_bo_get_offset(struct gbm_bo* bo, int plane)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
if (plane < 0 || plane >= bo->Impl.PlaneCount)
{
errno = EINVAL;
return 0;
}
return bo->Impl.Offsets[plane];
}
extern "C" struct gbm_device* gbm_bo_get_device(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
return bo->Impl.Owner;
}
extern "C" union gbm_bo_handle gbm_bo_get_handle(struct gbm_bo* bo)
{
union gbm_bo_handle handle;
handle.ptr = nullptr;
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
handle.s32 = -1;
return handle;
}
std::lock_guard<std::mutex> guard(*mutex);
// The kernel-level name of a dma-buf in this process is its file descriptor. A
// GEM handle would be a name that exists only relative to a DRM node, and this
// library deliberately does not require one, so the fd is the only answer here
// that is true for every buffer it hands out. Callers that need a name for a
// device should ask for the fd and pass that.
handle.s32 = bo->Impl.Fds[0];
return handle;
}
extern "C" int gbm_bo_get_fd(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return -1;
}
std::lock_guard<std::mutex> guard(*mutex);
// A dup, close-on-exec, owned by the caller. Closing it cannot affect the buffer
// or any other caller's copy.
return GbmBufferDupFd(*bo, 0);
}
extern "C" uint64_t gbm_bo_get_modifier(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return DrmFormatModInvalid;
}
std::lock_guard<std::mutex> guard(*mutex);
return bo->Impl.Modifier;
}
extern "C" int gbm_bo_get_plane_count(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
return bo->Impl.PlaneCount;
}
extern "C" union gbm_bo_handle gbm_bo_get_handle_for_plane(struct gbm_bo* bo, int plane)
{
union gbm_bo_handle handle;
handle.ptr = nullptr;
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
handle.s32 = -1;
return handle;
}
std::lock_guard<std::mutex> guard(*mutex);
if (plane < 0 || plane >= bo->Impl.PlaneCount)
{
errno = EINVAL;
handle.s32 = -1;
return handle;
}
// Same reasoning as gbm_bo_get_handle: the fd is the buffer's kernel name. Every
// plane of a self-allocated buffer lives in the one dma-buf, so all of them
// answer with that descriptor; the plane's position inside it is
// gbm_bo_get_offset.
const Int fd = bo->Impl.Fds[plane] >= 0 ? bo->Impl.Fds[plane] : bo->Impl.Fds[0];
handle.s32 = fd;
return handle;
}
extern "C" int gbm_bo_get_fd_for_plane(struct gbm_bo* bo, int plane)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return -1;
}
std::lock_guard<std::mutex> guard(*mutex);
return GbmBufferDupFd(*bo, plane);
}
extern "C" int gbm_bo_write(struct gbm_bo* bo, const void* buf, size_t count)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr || buf == nullptr)
{
errno = EINVAL;
return -1;
}
std::lock_guard<std::mutex> guard(*mutex);
if (count == 0)
{
return 0;
}
if (count > bo->Impl.AllocationSize)
{
// Refused rather than truncated: a caller that writes 4 MiB into a 1 MiB
// buffer and is told "0 bytes written, success" would read back memory it
// never wrote and call the result correct.
errno = EINVAL;
return -1;
}
// The write goes through the same counted mapping the rest of the API uses, so a
// caller that already holds the buffer mapped does not get a second mapping (or
// lose its own) behind its back.
Uint32 stride = 0;
void* address = GbmBufferMap(*bo, 0, 0, stride);
if (address == nullptr)
{
return -1;
}
memcpy(address, buf, count);
GbmBufferUnmap(*bo, address);
return 0;
}
extern "C" void gbm_bo_set_user_data(struct gbm_bo* bo, void* data,
void (*destroy_user_data)(struct gbm_bo*, void*))
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
return;
}
std::lock_guard<std::mutex> guard(*mutex);
// Assigning over existing data does not call the old callback: the record calls
// it exactly once, when the buffer goes away, which is the one moment a caller
// can still recognise the pointer it registered.
bo->Impl.UserData = data;
bo->Impl.UserDataDestroy = destroy_user_data;
}
extern "C" void* gbm_bo_get_user_data(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
return bo->Impl.UserData;
}
extern "C" void gbm_bo_destroy(struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(bo);
if (mutex == nullptr)
{
return;
}
std::lock_guard<std::mutex> guard(*mutex);
// Drops this caller's reference. A buffer a surface still holds stays alive -
// that is what the surface's own reference is for - so destroying a buffer that
// is currently presented is not a use-after-free waiting to happen.
GbmBufferRelease(*bo->Impl.Owner, *bo);
}
// ---------------------------------------------------------------------------
// Surfaces
// ---------------------------------------------------------------------------
extern "C" struct gbm_surface* gbm_surface_create(struct gbm_device* gbm, uint32_t width, uint32_t height,
uint32_t format, uint32_t flags)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
return GbmSurfaceCreate(*gbm, width, height, format, flags, nullptr, 0);
}
extern "C" struct gbm_surface* gbm_surface_create_with_modifiers(struct gbm_device* gbm, uint32_t width,
uint32_t height, uint32_t format,
const uint64_t* modifiers,
const unsigned int count)
{
// As with the buffer pair: the modifier-less form is the flags-0 case.
return gbm_surface_create_with_modifiers2(gbm, width, height, format, modifiers, count, 0);
}
extern "C" struct gbm_surface* gbm_surface_create_with_modifiers2(struct gbm_device* gbm, uint32_t width,
uint32_t height, uint32_t format,
const uint64_t* modifiers,
const unsigned int count, uint32_t flags)
{
std::mutex* mutex = MutexOf(gbm);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
return GbmSurfaceCreate(*gbm, width, height, format, flags, modifiers, (Uint32)count);
}
extern "C" struct gbm_bo* gbm_surface_lock_front_buffer(struct gbm_surface* surface)
{
std::mutex* mutex = MutexOf(surface);
if (mutex == nullptr)
{
errno = EINVAL;
return nullptr;
}
std::lock_guard<std::mutex> guard(*mutex);
// Null with errno = EBUSY when the compositor still holds every buffer: that is a
// state the caller has to handle, and pretending otherwise by handing out a
// buffer that is still being scanned out would be a torn frame rather than an
// error.
return GbmSurfaceLockFront(*surface);
}
extern "C" void gbm_surface_release_buffer(struct gbm_surface* surface, struct gbm_bo* bo)
{
std::mutex* mutex = MutexOf(surface);
if (mutex == nullptr || bo == nullptr)
{
return;
}
std::lock_guard<std::mutex> guard(*mutex);
GbmSurfaceReleaseBuffer(*surface, *bo);
}
extern "C" int gbm_surface_has_free_buffers(struct gbm_surface* surface)
{
std::mutex* mutex = MutexOf(surface);
if (mutex == nullptr)
{
errno = EINVAL;
return 0;
}
std::lock_guard<std::mutex> guard(*mutex);
// Truthful by construction: the swapchain exists from the moment the surface
// does, so this answers about buffers that were really allocated rather than
// about how many could still be.
return GbmSurfaceHasFreeBuffers(*surface) ? 1 : 0;
}
extern "C" void gbm_surface_destroy(struct gbm_surface* surface)
{
std::mutex* mutex = MutexOf(surface);
if (mutex == nullptr)
{
return;
}
std::lock_guard<std::mutex> guard(*mutex);
GbmSurfaceDestroy(*surface);
}
// ---------------------------------------------------------------------------
// Format names
// ---------------------------------------------------------------------------
extern "C" char* gbm_format_get_name(uint32_t gbm_format, struct gbm_format_name_desc* desc)
{
if (desc == nullptr)
{
errno = EINVAL;
return nullptr;
}
// The name of a fourcc is its four characters, most significant byte last, and
// that is all there is to it: a caller compares these strings against "AR24" and
// friends, and the big-endian bit is part of the value rather than something to
// spell out here.
desc->name[0] = (char)(gbm_format & 0xffu);
desc->name[1] = (char)((gbm_format >> 8) & 0xffu);
desc->name[2] = (char)((gbm_format >> 16) & 0xffu);
desc->name[3] = (char)((gbm_format >> 24) & 0xffu);
desc->name[4] = '\0';
return desc->name;
}
// End of Source File Header
File diff suppressed because it is too large Load Diff
+835
View File
@@ -0,0 +1,835 @@
// MobileGL - MobileGL/MG_Gbm/GbmHostChannel.cpp
// SPDX-License-Identifier: LGPL-3.0-only
//
// The host-frame backend: the socket the container receives the phone's frames on.
//
// The container half of MobileGL runs a Wayland compositor, and the frames that
// compositor renders into have to be the frames the phone already owns - the phone
// allocates them, its GPU draws into them, and its display scans them out. A buffer
// allocated from the container's own dma-heap would be a copy that has to travel back
// to the phone before anything can be shown, so when the display host is reachable this
// backend hands the compositor the phone's own buffers instead, and the fd the
// compositor passes on is the fd of a frame the display path already knows.
//
// What that costs is the one thing GBM normally promises: the geometry. A frame's
// width, height, stride and format are the display's and not the caller's, so a
// gbm_bo_create that takes an offered frame answers with the offer's geometry, and the
// caller reads it back with gbm_bo_get_width and its neighbours. That is this backend's
// documented behaviour rather than a surprise: the host decides what a frame is.
//
// Three rules keep this backend from ever being worse than not having it.
//
// * Nothing on the allocation path blocks. The socket is non-blocking and connect()
// to a pathname socket either succeeds or fails immediately, so a call that cannot
// be served from the host falls through to a local allocation inside the same call
// instead of waiting for a phone that may never answer. The one wait in this file is
// in device creation, after a display host has already accepted the connection, and
// it exists because of a measured race: the compositor builds its swapchain within
// microseconds of creating the device, while the offer for the first frame needs the
// round trip of the connection plus whatever the phone does before it sends. Without
// that wait the swapchain - the buffers that are shown for the life of the process -
// is allocated locally and the host's frames arrive for nobody to take.
// * A channel that dies is closed and marked down, and the next frame is allocated
// locally again. This library is loaded by a compositor that has to keep painting,
// so losing the display host may cost the frames their path to the screen but it
// may not cost the session.
// * A frame is released exactly once, when the record holding it is destroyed, and
// only into the connection that offered it. Releasing on gbm_bo_destroy would be a
// lie - the surface may still be scanning that frame out - and releasing an old
// frame into a new connection would free a slot belonging to another session.
//
// The device mutex is the only lock in here. Both callers of this file - the allocation
// path and the buffer record's destructor - already run under it, so there is no second
// lock to order against and no state of this file's own that another thread can see.
#include "GbmInternal.h"
#include <fcntl.h>
#include <poll.h>
#include <sys/socket.h>
#include <sys/uio.h>
#include <sys/un.h>
#include <unistd.h>
#include <cerrno>
#include <chrono>
namespace MobileGL::MG_Gbm
{
namespace
{
// One frame per message, and a message is the offer followed by the handle the
// offer describes. 4 KiB is far more than an AHardwareBuffer handle needs (a
// handle is a version, two counts and their two arrays) while still being small
// enough that a message which does not fit is a peer speaking something else -
// which is worth refusing rather than growing a buffer for.
constexpr SizeT GbmHostMessageMax = 4096;
// The descriptors one handle may carry. An AHardwareBuffer handle carries one
// (the dma-buf) in every layout this library has seen and the protocol reads only
// the first, so this sizes the control buffer rather than claiming anything about
// real handles.
constexpr Int GbmHostMaxFds = 8;
// How many messages one gbm_bo_create may look at before it gives up and
// allocates locally. An unusable message is skipped rather than fatal, so this
// bounds the work a peer that sends nothing but unusable messages can cause
// inside a single allocation.
constexpr Int GbmHostTakeAttempts = 16;
// The floor between two connection attempts. A container with no display host is
// the normal case, so a retry may not turn every allocation into a connect plus a
// log line; one attempt per second is often enough that a phone which comes up is
// picked up within a frame or two of the compositor's next allocation.
constexpr std::chrono::milliseconds GbmHostRetryInterval{1000};
// How long device creation waits for the display host's first frame once the
// connection is up. A frame the phone offers immediately needs one round trip on
// a unix socket, so this is orders of magnitude more than that and it is only
// ever spent by a device that has a host which then says nothing; a compile or a
// test run without a display host never reaches it, because the connect fails
// first.
constexpr Int GbmHostFirstOfferWaitMs = 250;
// Frame releases the socket would not take. A release is twelve bytes, so a full
// socket buffer is hundreds of frames' worth of them and this queue is only
// reached once the phone has stopped reading; the cap is what stops a peer that
// never reads from growing it without bound.
constexpr SizeT GbmHostMaxPendingReleases = 256;
// sun_path is where a pathname socket's name lives, and a name that does not fit
// is one the kernel cannot be told about at all.
constexpr SizeT GbmHostPathMax = sizeof(((struct sockaddr_un*)nullptr)->sun_path);
// close() may overwrite errno, and every caller here is in the middle of
// reporting a failure that errno describes. Saving it around the close keeps the
// reported failure the one that happened.
void GbmHostCloseQuietly(Int fd) noexcept
{
if (fd < 0)
{
return;
}
const Int savedErrno = errno;
close(fd);
errno = savedErrno;
}
} // namespace
// One connection to the display host, for the life of one device.
class GbmHostChannel
{
public:
explicit GbmHostChannel(const String& path) noexcept : Path(path) {}
~GbmHostChannel() noexcept;
GbmHostChannel(const GbmHostChannel&) = delete;
GbmHostChannel& operator=(const GbmHostChannel&) = delete;
// Connects when there is no connection and the retry interval has passed. A
// failure is never the caller's problem: the channel stays closed and the
// allocation that asked for a frame falls back to local memory.
void EnsureOpen(::gbm_device& device) noexcept;
// Waits, once, for the connection's first frame to be readable. Called from
// device creation and nowhere else: see the banner for why that one wait exists.
void AwaitFirstOffer(::gbm_device& device) noexcept;
// The next frame the host offered, or false when none is waiting.
Bool TakeFrame(::gbm_device& device, GbmHostFrame& frameOut) noexcept;
// Queues the release for one frame and tries to send what is queued.
void ReleaseFrame(::gbm_device& device, Uint32 seq, Uint64 generation) noexcept;
private:
// Turns one message into a frame. Everything it refuses is reported with the
// reason, because a peer this library cannot understand is the one thing that
// would otherwise be invisible from the outside.
Bool ParseOffer(const Uint8* payload, SizeT length, const Int* fds, Int fdCount,
GbmHostFrame& frameOut) noexcept;
// Sends what is queued. False means the connection can no longer be trusted to
// carry a twelve byte message, with errno describing what happened.
Bool FlushReleases() noexcept;
// Closes the socket and drops whatever was queued for it.
void CloseSocket() noexcept;
// Closes the socket and tells the device what that means for its backend name.
void MarkDown(::gbm_device& device, const char* what, Int error) noexcept;
String Path;
Int SocketFd = -1;
// Set by a connection this process has just made, and read once by the allocation that
// follows it: the accept and the frame that answers the dial happen on the other side of
// this socket, so an allocation that reads immediately after connecting reads before the
// answer to its own connection has been written. See TakeFrame.
Bool FreshConnection = false;
// Bumped by every connection, so a frame taken from a connection which has since
// died can be recognised as one that must not be released into its replacement.
Uint64 Generation = 0;
std::chrono::steady_clock::time_point NextRetry{};
// The errno of the last failed attempt, so that a failure which repeats once a
// second is not logged once a second.
Int LastFailureErrno = 0;
std::vector<GbmFrameRelease> PendingReleases;
Bool ReleaseOverflowLogged = false;
};
// ---- the offered-frame record ------------------------------------------
GbmHostFrame::~GbmHostFrame()
{
// The record owns the dup it was built on, which is why it is move-only: a copy
// would close the same descriptor twice.
if (Fd >= 0)
{
close(Fd);
Fd = -1;
}
}
GbmHostFrame::GbmHostFrame(GbmHostFrame&& other) noexcept
{
*this = static_cast<GbmHostFrame&&>(other);
}
GbmHostFrame& GbmHostFrame::operator=(GbmHostFrame&& other) noexcept
{
if (this == &other)
{
return *this;
}
if (Fd >= 0)
{
close(Fd);
}
Seq = other.Seq;
Index = other.Index;
Width = other.Width;
Height = other.Height;
StrideBytes = other.StrideBytes;
Format = other.Format;
BytesPerPixel = other.BytesPerPixel;
Usage = other.Usage;
Generation = other.Generation;
Fd = other.Fd;
Size = other.Size;
// The source must not close what this record now owns.
other.Fd = -1;
return *this;
}
// ---- the connection ----------------------------------------------------
GbmHostChannel::~GbmHostChannel() noexcept
{
// One last try for whatever is queued - the phone may still be reading - and
// then the socket goes away with the process.
if (SocketFd >= 0 && !PendingReleases.empty())
{
FlushReleases();
}
if (!PendingReleases.empty())
{
MGLOG_W("gbm: %zu frame release(s) could not be sent to the display host at %s; the phone reclaims "
"those slots when the session ends",
PendingReleases.size(), Path.c_str());
}
CloseSocket();
}
void GbmHostChannel::EnsureOpen(::gbm_device& device) noexcept
{
if (SocketFd >= 0)
{
return;
}
const std::chrono::steady_clock::time_point now = std::chrono::steady_clock::now();
if (now < NextRetry)
{
return;
}
NextRetry = now + GbmHostRetryInterval;
// SOCK_SEQPACKET because a frame and its descriptors are one message and the
// reader may not reassemble either half; SOCK_NONBLOCK because the caller is a
// compositor's frame loop and may not be parked on a phone; SOCK_CLOEXEC because
// every process this compositor starts would otherwise inherit a socket it can
// read frames from.
const Int fd = socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0);
if (fd < 0)
{
MGLOG_E_ONCE("gbm: cannot create the socket to the display host (%d: %s); every frame is allocated "
"locally",
errno, strerror(errno));
return;
}
if (Path.empty() || Path.size() >= GbmHostPathMax)
{
MGLOG_E_ONCE("gbm: the display host socket path \"%s\" does not fit in sun_path (%zu bytes); every "
"frame is allocated locally",
Path.c_str(), GbmHostPathMax);
GbmHostCloseQuietly(fd);
return;
}
struct sockaddr_un address = {};
address.sun_family = AF_UNIX;
// The terminator travels with the name: sun_path is a C string for a pathname
// socket, and a name without one is a name the kernel reads past the end of.
memcpy(address.sun_path, Path.c_str(), Path.size() + 1);
if (connect(fd, reinterpret_cast<struct sockaddr*>(&address), sizeof(address)) != 0)
{
const Int connectErrno = errno;
GbmHostCloseQuietly(fd);
// A failure that repeats once a second is not news: a container with no
// display host fails here on every retry, and a line per second saying so
// would bury the rest of the log. A failure that changed is news, because it
// says the path or the peer changed.
if (connectErrno != LastFailureErrno)
{
LastFailureErrno = connectErrno;
if (connectErrno == EPROTOTYPE)
{
MGLOG_E("gbm: %s is a socket of a different type (%d: %s). The frame protocol is "
"SOCK_SEQPACKET - one message per frame, carrying that frame's descriptors - "
"because a reader cannot tell from a byte stream which frame a descriptor "
"belongs to; frames are allocated locally instead",
Path.c_str(), connectErrno, strerror(connectErrno));
}
else
{
MGLOG_I("gbm: no display host at %s (%d: %s); every frame is allocated locally until one "
"answers",
Path.c_str(), connectErrno, strerror(connectErrno));
}
}
return;
}
SocketFd = fd;
FreshConnection = true;
// A new connection is a new session with its own frames, which is what makes the
// generation a correct answer to "may this release be sent here".
Generation++;
LastFailureErrno = 0;
device.Impl.HostAvailable = true;
device.Impl.BackendName = "mobilegl-host-frame";
MGLOG_I("gbm: connected to the display host at %s (one SOCK_SEQPACKET message per frame, up to %zu "
"bytes); frames offered over it become gbm_bos whose geometry is the offer's",
Path.c_str(), GbmHostMessageMax);
}
void GbmHostChannel::AwaitFirstOffer(::gbm_device& device) noexcept
{
if (SocketFd < 0)
{
return;
}
const std::chrono::steady_clock::time_point deadline =
std::chrono::steady_clock::now() + std::chrono::milliseconds(GbmHostFirstOfferWaitMs);
for (;;)
{
const std::chrono::milliseconds left =
std::chrono::duration_cast<std::chrono::milliseconds>(deadline - std::chrono::steady_clock::now());
if (left.count() <= 0)
{
break;
}
// poll() and not recvmsg(): the frame has to stay in the socket for the
// allocation that is about to ask for it, and a peek at a message carrying
// SCM_RIGHTS is a question with no safe answer.
struct pollfd waiting = {};
waiting.fd = SocketFd;
waiting.events = POLLIN;
const Int ready = poll(&waiting, 1, (Int)left.count());
if (ready > 0)
{
MGLOG_I("gbm: the display host at %s has a frame waiting; the buffers the compositor is about "
"to allocate come from it",
Path.c_str());
return;
}
if (ready == 0)
{
break;
}
if (errno != EINTR)
{
MarkDown(device, "poll", errno);
return;
}
}
MGLOG_W("gbm: the display host at %s accepted the connection but offered no frame within %d ms; the "
"buffers allocated before its first frame are local ones",
Path.c_str(), GbmHostFirstOfferWaitMs);
}
Bool GbmHostChannel::TakeFrame(::gbm_device& device, GbmHostFrame& frameOut) noexcept
{
EnsureOpen(device);
if (SocketFd < 0)
{
return false;
}
// THE FIRST ALLOCATION AFTER A DIAL WAITS FOR THE ANSWER TO IT. The accept and the offer that
// answers the connection this call has just made are written by the phone, a round trip
// behind the connect() that returns here — and what is allocated before that offer arrives
// is a local buffer, which is what a compositor builds its whole swapchain out of and then
// keeps for the life of the process. So a connection made on this path is given the same
// first-offer window device creation gives one, once, and only while it is new.
if (FreshConnection)
{
FreshConnection = false;
AwaitFirstOffer(device);
}
// Releases the socket would not take last time are retried here, because this is
// the one call in the library that happens once per frame and therefore the
// natural place for a queue waiting for room to get its next chance.
if (!PendingReleases.empty() && !FlushReleases())
{
MarkDown(device, "sendmsg", errno);
return false;
}
Uint8 payload[GbmHostMessageMax];
for (Int attempt = 0; attempt < GbmHostTakeAttempts; ++attempt)
{
Int receivedFds[GbmHostMaxFds];
for (Int i = 0; i < GbmHostMaxFds; ++i)
{
receivedFds[i] = -1;
}
struct iovec payloadVector = {};
payloadVector.iov_base = payload;
payloadVector.iov_len = sizeof(payload);
Uint8 control[CMSG_SPACE(sizeof(Int) * GbmHostMaxFds)] = {};
struct msghdr message = {};
message.msg_iov = &payloadVector;
message.msg_iovlen = 1;
message.msg_control = control;
message.msg_controllen = sizeof(control);
// MSG_DONTWAIT for the promise at the top of this file, and MSG_CMSG_CLOEXEC
// so a descriptor the phone sent does not survive into a child of this
// compositor, which would then hold a frame nothing will ever release.
const ssize_t received = recvmsg(SocketFd, &message, MSG_DONTWAIT | MSG_CMSG_CLOEXEC);
if (received < 0)
{
if (errno == EAGAIN || errno == EWOULDBLOCK)
{
// Nothing offered yet, which is what an idle display looks like. The
// caller allocates locally and asks again on its next frame.
return false;
}
if (errno == EINTR)
{
continue;
}
MarkDown(device, "recvmsg", errno);
return false;
}
// The descriptors come out of the control data before anything is decided
// about the message, because they belong to this process from the moment it
// was received: a path that returns without closing them leaks a descriptor
// per unusable message.
Int fdCount = 0;
for (struct cmsghdr* header = CMSG_FIRSTHDR(&message); header != nullptr;
header = CMSG_NXTHDR(&message, header))
{
if (header->cmsg_level != SOL_SOCKET || header->cmsg_type != SCM_RIGHTS)
{
continue;
}
const SizeT bytes = (SizeT)header->cmsg_len - CMSG_LEN(0);
const Int count = (Int)(bytes / sizeof(Int));
const Int* descriptors = reinterpret_cast<const Int*>(CMSG_DATA(header));
for (Int i = 0; i < count; ++i)
{
if (fdCount < GbmHostMaxFds)
{
receivedFds[fdCount++] = descriptors[i];
}
else
{
// More descriptors than a frame can carry: they are this
// process's now, so they are closed here rather than leaked.
GbmHostCloseQuietly(descriptors[i]);
}
}
}
if ((message.msg_flags & MSG_CTRUNC) != 0)
{
MGLOG_W_ONCE("gbm: an offered frame's control data did not fit in %zu bytes; the frame is "
"skipped because its descriptor list is incomplete",
sizeof(control));
for (Int i = 0; i < fdCount; ++i)
{
GbmHostCloseQuietly(receivedFds[i]);
}
continue;
}
if (received == 0)
{
// Zero bytes on a connected socket is the end of it, whether the phone
// closed cleanly or the kernel gave up on it. Either way the frames have
// to come from somewhere else from here on.
for (Int i = 0; i < fdCount; ++i)
{
GbmHostCloseQuietly(receivedFds[i]);
}
MarkDown(device, "the display host closed the connection", 0);
return false;
}
if ((message.msg_flags & MSG_TRUNC) != 0)
{
// Longer than a frame message can be, so it is not one - and on a
// message-oriented socket the rest is already gone while the next message
// is still a whole message, which is why this skips one message instead
// of ending the channel.
MGLOG_W_ONCE("gbm: the display host sent a message longer than %zu bytes; it is skipped",
GbmHostMessageMax);
for (Int i = 0; i < fdCount; ++i)
{
GbmHostCloseQuietly(receivedFds[i]);
}
continue;
}
const Bool parsed = ParseOffer(payload, (SizeT)received, receivedFds, fdCount, frameOut);
// Whatever the frame did not keep is closed here: the copy of the dma-buf
// descriptor the phone sent (the frame owns its own dup) and the rest of the
// handle, which is a layout this library does not interpret.
for (Int i = 0; i < fdCount; ++i)
{
GbmHostCloseQuietly(receivedFds[i]);
}
if (parsed)
{
return true;
}
}
return false;
}
Bool GbmHostChannel::ParseOffer(const Uint8* payload, SizeT length, const Int* fds, Int fdCount,
GbmHostFrame& frameOut) noexcept
{
constexpr SizeT HeaderBytes = sizeof(GbmFrameOffer) + sizeof(AHBNativeHandleHeader);
if (length < HeaderBytes)
{
MGLOG_W_ONCE("gbm: the display host sent a %zu byte message where a frame message is at least %zu "
"bytes (the offer followed by the handle header); it is skipped",
length, HeaderBytes);
return false;
}
GbmFrameOffer offer = {};
memcpy(&offer, payload, sizeof(offer));
if (offer.Magic != GbmFrameOfferMagic)
{
MGLOG_W_ONCE("gbm: a message from the display host begins with 0x%08x rather than 'MGGF'; it is "
"skipped",
offer.Magic);
return false;
}
if (offer.Version != GbmHostProtocolVersion)
{
// Refused rather than read as if the versions agreed: the version is what
// says how the rest of the message is laid out, and a frame read with the
// wrong layout is memory the compositor would render into.
MGLOG_E_ONCE("gbm: the display host offers frames in protocol version %u where this library speaks "
"version %u; its frames are refused until the two agree",
offer.Version, (Uint32)GbmHostProtocolVersion);
return false;
}
AHBNativeHandleHeader handle = {};
memcpy(&handle, payload + sizeof(offer), sizeof(handle));
if (handle.NumFds < 1 || handle.NumInts < 0)
{
MGLOG_W_ONCE("gbm: an offered frame's handle carries %d descriptor(s) and %d integer(s), where the "
"dma-buf is the first descriptor and there has to be one; the frame is skipped",
handle.NumFds, handle.NumInts);
return false;
}
// The handle describes itself with its two counts, so the message has to be
// exactly as long as those counts make it. Any other length means the two ends
// disagree about the layout, and a frame built from a message whose layout was
// guessed is a buffer of unknown memory.
const SizeT expected = HeaderBytes + ((SizeT)handle.NumFds + (SizeT)handle.NumInts) * sizeof(Int32);
if (length != expected)
{
MGLOG_W_ONCE("gbm: an offered frame's message is %zu bytes where its handle (%d descriptors, %d "
"integers) makes it %zu; the frame is skipped",
length, handle.NumFds, handle.NumInts, expected);
return false;
}
if (fdCount < handle.NumFds)
{
MGLOG_W_ONCE("gbm: an offered frame's handle names %d descriptor(s) and %d arrived with the "
"message; the frame is skipped",
handle.NumFds, fdCount);
return false;
}
if (offer.Width == 0 || offer.Height == 0 || offer.StrideBytes == 0)
{
MGLOG_W_ONCE("gbm: an offered frame is %ux%u with a stride of %u bytes; a buffer with no size is "
"not one this library can hand out",
offer.Width, offer.Height, offer.StrideBytes);
return false;
}
// The frame is built on a dup of the first descriptor rather than on the
// descriptor itself: the copy that arrived belongs to the message, and the frame
// outlives it. Close-on-exec, like every other descriptor this library hands out.
const Int frameFd = fcntl(fds[0], F_DUPFD_CLOEXEC, 0);
if (frameFd < 0)
{
MGLOG_E_ONCE("gbm: cannot dup the descriptor of an offered frame (%d: %s)", errno, strerror(errno));
return false;
}
// How long the frame really is, asked of the descriptor instead of computed from
// the geometry: a gralloc dma-buf is usually larger than stride x height, and a
// mapping has to cover what exists rather than what the arithmetic expects.
SizeT size = 0;
const off_t end = lseek(frameFd, 0, SEEK_END);
if (end > 0)
{
size = (SizeT)end;
}
const SizeT described = (SizeT)offer.StrideBytes * (SizeT)offer.Height;
if (size == 0)
{
// A descriptor that does not answer for its own length - an exporter that
// does not implement llseek - leaves the offer's arithmetic as the only
// length there is.
size = described;
}
if (size < described)
{
MGLOG_W_ONCE("gbm: an offered frame's descriptor holds %zu bytes where %u rows of %u bytes need %zu; "
"the frame is skipped rather than mapped over memory that is not there",
size, offer.Height, offer.StrideBytes, described);
GbmHostCloseQuietly(frameFd);
return false;
}
frameOut.Seq = offer.Seq;
frameOut.Index = offer.Index;
frameOut.Width = offer.Width;
frameOut.Height = offer.Height;
frameOut.StrideBytes = offer.StrideBytes;
frameOut.Format = offer.Format;
frameOut.Usage = ((Uint64)offer.UsageHi << 32) | (Uint64)offer.UsageLo;
frameOut.Generation = Generation;
frameOut.Fd = frameFd;
frameOut.Size = size;
frameOut.BytesPerPixel = GbmFormatBytesPerPixel(offer.Format);
if (frameOut.BytesPerPixel == 0)
{
// A format this library has no table entry for is still a frame the phone
// drew. The offer's stride is then the only description of a pixel that
// exists: a row of Width pixels is StrideBytes long.
frameOut.BytesPerPixel = offer.StrideBytes / offer.Width;
}
MGLOG_I("gbm: frame seq=%u index=%u %ux%u stride=%u format=0x%08x (%u bytes per pixel, %zu bytes "
"measured) taken from the display host",
offer.Seq, offer.Index, offer.Width, offer.Height, offer.StrideBytes, offer.Format,
frameOut.BytesPerPixel, size);
return true;
}
void GbmHostChannel::ReleaseFrame(::gbm_device& device, Uint32 seq, Uint64 generation) noexcept
{
if (SocketFd < 0)
{
// A release only means anything on the connection that offered the frame,
// and there is none.
return;
}
if (generation != Generation)
{
// The frame was offered by a connection that has since died and been
// replaced. Sending this into the replacement would name a frame of another
// session, and the phone would recycle a slot this buffer was not made of.
MGLOG_W_ONCE("gbm: a frame offered by a display host connection that has since died is not "
"released into its replacement");
return;
}
if (PendingReleases.size() >= GbmHostMaxPendingReleases)
{
if (!ReleaseOverflowLogged)
{
ReleaseOverflowLogged = true;
MGLOG_W("gbm: %zu frame releases to the display host at %s are already waiting for room in the "
"socket; this one is dropped rather than queued without bound, and the phone reclaims "
"the slot when the session ends",
PendingReleases.size(), Path.c_str());
}
return;
}
GbmFrameRelease release = {};
release.Magic = GbmFrameReleaseMagic;
release.Version = GbmHostProtocolVersion;
release.Seq = seq;
PendingReleases.push_back(release);
if (!FlushReleases())
{
MarkDown(device, "sendmsg", errno);
}
}
Bool GbmHostChannel::FlushReleases() noexcept
{
while (!PendingReleases.empty())
{
struct iovec releaseVector = {};
releaseVector.iov_base = &PendingReleases.front();
releaseVector.iov_len = sizeof(GbmFrameRelease);
struct msghdr message = {};
message.msg_iov = &releaseVector;
message.msg_iovlen = 1;
// MSG_DONTWAIT for the same reason every other call here is non-blocking, and
// MSG_NOSIGNAL because a compositor killed by SIGPIPE when the phone goes
// away is a desktop that disappears with it.
const ssize_t sent = sendmsg(SocketFd, &message, MSG_DONTWAIT | MSG_NOSIGNAL);
if (sent == (ssize_t)sizeof(GbmFrameRelease))
{
PendingReleases.erase(PendingReleases.begin());
continue;
}
if (sent < 0 && errno == EINTR)
{
continue;
}
if (sent < 0 && (errno == EAGAIN || errno == EWOULDBLOCK))
{
// A full socket is a phone that has stopped reading rather than a dead
// one: the release stays queued and goes out with a later frame.
return true;
}
// A short send, a reset and a broken pipe all say the same thing: this
// connection can no longer be trusted to have carried a twelve byte message,
// and acting on a maybe is worse than handing the slot back when the session
// ends. errno describes which of them it was.
return false;
}
return true;
}
void GbmHostChannel::CloseSocket() noexcept
{
if (SocketFd >= 0)
{
GbmHostCloseQuietly(SocketFd);
SocketFd = -1;
}
PendingReleases.clear();
}
void GbmHostChannel::MarkDown(::gbm_device& device, const char* what, Int error) noexcept
{
if (!PendingReleases.empty())
{
MGLOG_W("gbm: %zu frame release(s) were still queued for the display host and cannot be sent over a "
"connection that is gone; the phone reclaims those slots when the session ends",
PendingReleases.size());
}
CloseSocket();
if (error != 0)
{
MGLOG_W("gbm: the display host connection at %s is down: %s failed (%d: %s); frames are allocated "
"locally from here on, and a new connection is attempted on a later allocation",
Path.c_str(), what, error, strerror(error));
}
else
{
MGLOG_W("gbm: the display host connection at %s is down: %s; frames are allocated locally from here "
"on, and a new connection is attempted on a later allocation",
Path.c_str(), what);
}
device.Impl.HostAvailable = false;
device.Impl.BackendName = device.Impl.SelfAllocatedBackendName;
}
// ---- the device-facing entry points ------------------------------------
void GbmHostChannelCreate(::gbm_device& device) noexcept
{
const char* configured = getenv("MOBILEGL_GBM_SOCKET");
const String path =
(configured != nullptr && configured[0] != '\0') ? String(configured) : String("/run/mobilegl-gbm.sock");
device.Impl.SocketPath = path;
try
{
GbmHostChannel* channel = new GbmHostChannel(path);
device.Impl.Host = channel;
// The one attempt made at creation. It is allowed to fail: a machine that is
// not the container half of a split run has no display host, and that is a
// device which allocates locally rather than a device that could not exist.
channel->EnsureOpen(device);
// A connection that came up is waited on once, and only here, so that the
// compositor's swapchain is built out of the host's frames rather than out of
// local ones it will keep for the life of the process.
channel->AwaitFirstOffer(device);
}
catch (...)
{
device.Impl.Host = nullptr;
device.Impl.HostAvailable = false;
MGLOG_W("gbm: the display host channel could not be created; every frame is allocated locally");
}
}
void GbmHostChannelDestroy(::gbm_device& device) noexcept
{
GbmHostChannel* channel = device.Impl.Host;
if (channel == nullptr)
{
return;
}
device.Impl.Host = nullptr;
device.Impl.HostAvailable = false;
delete channel;
}
Bool GbmHostChannelTakeFrame(::gbm_device& device, GbmHostFrame& frameOut) noexcept
{
GbmHostChannel* channel = device.Impl.Host;
if (channel == nullptr)
{
return false;
}
return channel->TakeFrame(device, frameOut);
}
void GbmHostChannelReleaseFrame(::gbm_device& device, Uint32 seq, Uint64 generation) noexcept
{
GbmHostChannel* channel = device.Impl.Host;
if (channel == nullptr)
{
return;
}
channel->ReleaseFrame(device, seq, generation);
}
} // namespace MobileGL::MG_Gbm
// End of Source File Header
+476
View File
@@ -0,0 +1,476 @@
// MobileGL - MobileGL/MG_Gbm/GbmInternal.h
// SPDX-License-Identifier: LGPL-3.0-only
//
// This header carries the internal model shared by the GBM front-end files: the
// device, buffer-object and surface records, the two allocation backends, and the
// wire protocol used to talk to a phone that owns the real display hardware.
//
// The reason a GBM implementation has to exist in this tree at all is that the
// container half of MobileGL runs a Wayland compositor (kwin). That compositor
// needs libgbm to allocate the buffers it renders into, and the fd of such a
// buffer is handed to the Anland display producer, which ships the frame to the
// phone. If Mesa's libgbm is loaded, the compositor renders into Mesa buffers and
// the phone ends up receiving a copy through a path that was never designed to
// carry them. When this library is loaded instead, the buffers are either frames
// the phone itself offered to us, or buffers allocated from the kernel's dma-heap
// so that they are dma-bufs the display host can consume directly.
//
// The ownership rules below are the whole contract, because every one of them is
// observable from outside this library:
//
// * A buffer record owns its descriptors. gbm_bo_get_fd and
// gbm_bo_get_fd_for_plane hand out a dup, so a caller that closes what it was
// given cannot invalidate the buffer, and the matching close happens exactly
// once, in ~GbmBufferObject.
// * A mapping is one mmap of the buffer's fd, counted rather than duplicated, so
// two overlapping gbm_bo_map calls share one address and the last matching
// gbm_bo_unmap releases it.
// * A surface owns one reference per swapchain buffer, and every
// gbm_surface_lock_front_buffer hands the caller a reference of its own. That
// is what makes gbm_bo_destroy on a locked buffer harmless: it drops the
// caller's reference and leaves the surface's bookkeeping intact.
// * One mutex per device. GBM calls arrive from the compositor's render thread
// and from its main thread, and the state they touch (the buffer table, the
// swapchains, the backends) is small enough that one lock is cheaper than
// several - and, more importantly, one lock cannot be taken in the wrong order.
//
// The helpers declared at the bottom of this file are the entire internal interface
// between GbmDevice.cpp (the records and the self-allocated backend) and GbmApi.cpp
// (the extern "C" surface). Every one of them assumes the device mutex is already
// held: the lock is taken once, in the API layer, so that a compositor call can
// never interleave with another half way through a swapchain update.
#pragma once
#include <Includes.h>
#include <gbm.h>
#include <atomic>
#include <cstdint>
#include <cstring>
#include <mutex>
#include <string>
#include <unordered_map>
#include <vector>
namespace MobileGL::MG_Gbm
{
// DRM_FORMAT_MOD_LINEAR and DRM_FORMAT_MOD_INVALID. drm_fourcc.h is not part of
// libgbm's interface, and these two values are frozen kernel UAPI, so they are
// spelled out here rather than made a build dependency on libdrm's headers.
static constexpr Uint64 DrmFormatModLinear = 0ULL;
static constexpr Uint64 DrmFormatModInvalid = 0x00ffffffffffffffULL;
// Every buffer this library hands out is described by one of these. HostFrame
// buffers are owned by the phone and only borrowed by us, so they must be
// released back over the channel instead of being freed locally.
enum class GbmBackend
{
None = 0,
HostFrame,
DmaHeap,
MemFd,
};
// The phone offers frames with this packed struct immediately followed by an
// AHardwareBuffer native handle in the payload of one message, with the handle's
// descriptors in the SCM_RIGHTS control data of that same message. The layout is
// fixed by the phone side, so the packing attribute is load bearing and the field
// order may not be rearranged.
//
// The transport below the two structs is SOCK_SEQPACKET, and that is part of the
// protocol rather than an implementation detail: a frame and its descriptors only
// mean anything together, and the message boundary is what lets the reader take
// the two as one thing. A stream socket would let a reader receive half an offer,
// or a descriptor whose position in the byte stream it had to reconstruct, and the
// failure mode of getting that arithmetic wrong is a frame silently made of
// another frame's memory.
//
// Format is a DRM fourcc and not an AHardwareBuffer format, because it is what
// gbm_bo_get_format has to answer with and a compositor compares it against
// DRM_FORMAT_* values. StrideBytes is the byte distance between two rows of plane
// 0, which is the same single-stride description GBM_BO_IMPORT_FD carries - and
// the reason a frame read from this struct is reported as linear.
struct GbmFrameOffer
{
uint32_t Magic; // 'M','G','G','F' little endian on both ends of the wire
uint32_t Version;
uint32_t Seq;
uint32_t Index;
uint32_t Width;
uint32_t Height;
uint32_t StrideBytes;
uint32_t Format;
uint32_t UsageLo;
uint32_t UsageHi;
} __attribute__((packed));
// The container replies with this once it no longer reads the frame, which is
// what lets the phone recycle the slot. Without it the display host would run
// out of buffers after a handful of frames.
struct GbmFrameRelease
{
uint32_t Magic; // 'M','G','G','R'
uint32_t Version;
uint32_t Seq;
} __attribute__((packed));
static constexpr uint32_t GbmFrameOfferMagic = 0x4647474Du; // "MGGF"
static constexpr uint32_t GbmFrameReleaseMagic = 0x5247474Du; // "MGGR"
static constexpr uint32_t GbmHostProtocolVersion = 1;
// THE REVERSE DIRECTION: this end asking the render server to back one of its textures with a
// dma-buf. The compositor's scene reaches the display daemon's buffers through
// glEGLImageTargetTexture2DOES, and the server has to be told which texture that dma-buf backs -
// over this same socket, because it is the one channel that can carry a descriptor across the
// container boundary. Restated here rather than shared with MG_Backend/GbmFrameChannel.h for the
// same reason the two frame shapes are: the ends are built into different libraries from
// different build configurations (that file is the Android render server and never sees libgbm).
// Field for field, packing included; the plane descriptors ride in the SCM_RIGHTS control data of
// the same sendmsg, one message per import.
struct GbmImportRequest
{
uint32_t Magic; // 'M','G','G','I'
uint32_t Version;
uint32_t Target; // the GL texture target the image was bound to (0x8D65 = GL_TEXTURE_EXTERNAL_OES)
uint32_t TexName; // THIS side's texture name; the server resolves its twin
uint32_t Width;
uint32_t Height;
uint32_t FourCC;
uint32_t PlaneCount;
int32_t Stride[4];
int32_t Offset[4];
uint64_t Modifier[4];
uint32_t HasModifier[4];
} __attribute__((packed));
struct GbmImportReply
{
uint32_t Magic; // 'I','G','G','R'
uint32_t Version;
uint32_t Ok;
uint32_t Reserved;
} __attribute__((packed));
static constexpr uint32_t GbmImportRequestMagic = 0x4947474Du; // "MGGI"
static constexpr uint32_t GbmImportReplyMagic = 0x52474749u; // "IGGR"
static constexpr uint32_t GbmImportMaxPlanes = 4;
// An AHardwareBuffer native handle as it appears on the wire: three counts
// followed by numFds file descriptors and numInts integers. The counts are
// 32 bit even on 64 bit Android, which is why they are spelled out here
// rather than using the platform's native_handle_t.
struct AHBNativeHandleHeader
{
int32_t Version;
int32_t NumFds;
int32_t NumInts;
} __attribute__((packed));
static constexpr uint32_t GbmMaxPlanes = 4;
// How one plane of a buffer is laid out inside the single allocation that backs
// it. Multi-plane formats keep every plane in one dma-buf and describe the
// separation with Offsets, which is the layout a modifier of
// DRM_FORMAT_MOD_LINEAR can express and the only one this library produces.
struct GbmPlaneLayout
{
Uint32 Stride = 0;
Uint32 Offset = 0;
SizeT Size = 0;
};
// One frame the phone offered, as this process holds it: the offer's own
// description of the memory, plus the descriptor this library duped out of the
// handle the phone sent. It owns that descriptor for as long as it lives, so a
// caller that takes a frame and then fails still closes it, and the descriptor the
// phone sent is closed with the rest of the message's descriptors rather than kept.
struct GbmHostFrame
{
Uint32 Seq = 0;
Uint32 Index = 0;
Uint32 Width = 0;
Uint32 Height = 0;
Uint32 StrideBytes = 0;
Uint32 Format = 0;
// Bytes per pixel of plane 0. The offer carries a fourcc, so it comes from the
// format table when there is an entry for it and from the stride otherwise.
Uint32 BytesPerPixel = 0;
Uint64 Usage = 0;
// Which connection offered the frame. A release names a frame, and a frame only
// means anything to the connection that offered it.
Uint64 Generation = 0;
// The duped dma-buf descriptor, and how many bytes it actually backs as asked
// of the descriptor itself rather than computed from the offered geometry.
Int Fd = -1;
SizeT Size = 0;
// Move-only, because two records naming one descriptor would close it twice.
GbmHostFrame() = default;
GbmHostFrame(const GbmHostFrame&) = delete;
GbmHostFrame& operator=(const GbmHostFrame&) = delete;
GbmHostFrame(GbmHostFrame&& other) noexcept;
GbmHostFrame& operator=(GbmHostFrame&& other) noexcept;
~GbmHostFrame();
};
// One imported or locally allocated buffer. The struct is private to this
// library; gbm_bo is the public opaque name and is defined in terms of it.
struct GbmBufferObject
{
Uint32 Width = 0;
Uint32 Height = 0;
Uint32 Format = 0;
Uint32 Stride = 0;
Uint64 Modifier = DrmFormatModLinear;
Uint64 Usage = 0;
Int PlaneCount = 1;
// Bytes per pixel of plane 0. Kept beside the layout because gbm_bo_map
// needs it to place a sub-rectangle, and because a format this library can
// allocate but cannot describe in bytes would be a format it must refuse.
Uint32 BytesPerPixel = 0;
// Fds owned by this record. get_fd and get_fd_for_plane hand out dups so
// that a caller closing its fd can never invalidate the buffer. Index 0 is
// the only one a self-allocated buffer has: every plane of a linear image
// lives in that one dma-buf, and Fds[1..] stay -1 for it. An import with
// several descriptors fills several entries, and the destructor closes
// every entry that is not -1.
Int Fds[GbmMaxPlanes] = {-1, -1, -1, -1};
Uint32 Strides[GbmMaxPlanes] = {0, 0, 0, 0};
Uint32 Offsets[GbmMaxPlanes] = {0, 0, 0, 0};
Uint64 Modifiers[GbmMaxPlanes] = {0, 0, 0, 0};
// The number of bytes the descriptors actually back, rounded up to the page
// size because that is what both backends allocate. It is the only length
// that is safe to mmap: a mapping longer than the object faults on the last
// page, and a shorter one would hide the caller's own stride arithmetic.
SizeT AllocationSize = 0;
GbmBackend Backend = GbmBackend::None;
// Set only for host frames: the slot index and sequence number that the
// release message has to mention so the phone can recycle the right slot, and
// the connection the frame arrived on. A frame left over from a connection
// that has since died is never released into the one that replaced it, because
// that would free a slot belonging to another session.
Uint32 HostIndex = 0;
Uint32 HostSeq = 0;
Uint64 HostGeneration = 0;
Bool HostOwned = false;
// How many maps are outstanding. The mapping itself is kept here so that
// unmap can find and tear down the right one.
Int MapRefCount = 0;
void* MapAddress = nullptr;
SizeT MapSize = 0;
// Surfaces keep their buffers alive by count rather than by ownership, so
// the device knows when a bo is truly unreferenced.
Int RefCount = 1;
// gbm_bo_set_user_data's pair. The destroy callback is invoked with the
// gbm_bo that is going away, once, from the destructor below.
void* UserData = nullptr;
void (*UserDataDestroy)(::gbm_bo*, void*) = nullptr;
::gbm_device* Owner = nullptr;
// The record releases its own descriptors and its own mapping, so that no
// failure path anywhere in this library can leak either one - an allocation
// that fails half way through construction still cleans up when the record
// it was building is destroyed.
~GbmBufferObject();
};
// A real swapchain. The compositor asks for a buffer to render into, presents
// it, and later returns it; the three states below are exactly the states the
// GBM API can express.
struct GbmSurface
{
Uint32 Width = 0;
Uint32 Height = 0;
Uint32 Format = 0;
Uint32 Flags = 0;
std::vector<::gbm_bo*> Buffers;
// Index of the buffer currently handed to the compositor as the front
// buffer, or -1 when nothing has been locked.
Int LockedIndex = -1;
// Indices of buffers the compositor has released and that may be handed
// out again, plus the one that is locked.
std::vector<Bool> Free;
::gbm_device* Owner = nullptr;
// Dropping the surface's references here is what keeps the creation path
// free of a second failure path: if the third swapchain buffer cannot be
// allocated, unwinding the half-built surface returns the two that exist.
~GbmSurface();
};
// The device ties one backend choice, one host channel and one mutex together.
// The single mutex is deliberate: GBM calls arrive from the compositor's
// render thread and from its main thread, and the state they touch (the bo
// table, the backends) is small enough that one lock is cheaper than several.
struct GbmDevice
{
// The fd this device was created with. It is a private dup taken at
// creation, never the caller's descriptor, so gbm_device_destroy can close
// it without ever closing an fd the caller still has open. -1 when the
// caller passed none and no render node could be opened: the self-allocated
// backend does not need a DRM node, so that is a working device and not an
// error.
Int DeviceFd = -1;
// The dma-heap node this device allocates from, or -1 when none could be
// opened, in which case every allocation falls back to a memfd.
Int HeapFd = -1;
String HeapPath;
// Where the display host is reached, as configured when the device was created.
String SocketPath;
GbmBackend DefaultBackend = GbmBackend::DmaHeap;
// True while the channel below is connected and frames are arriving over it.
// False means every allocation this device answers is answered locally.
Bool HostAvailable = false;
// The name gbm_device_get_backend_name reports. It names what this device
// actually does, because a compositor that logs it is asking exactly that - so
// it changes when the host channel connects and when that connection dies, and
// SelfAllocatedBackendName is what it falls back to when it does.
String BackendName = "mobilegl";
String SelfAllocatedBackendName = "mobilegl";
std::mutex Mutex;
std::unordered_map<::gbm_bo*, GbmBufferObject*> Buffers;
std::unordered_map<::gbm_surface*, GbmSurface*> Surfaces;
// The host channel is a separate translation unit so that the socket protocol
// never leaks into the allocation paths. It lives for as long as the device
// does and is reached only under this device's mutex, which is what lets the
// allocation path ask it for a frame without a second lock to order against.
class GbmHostChannel* Host = nullptr;
// Closes the heap node and the private device dup, so that a device which
// fails to finish constructing still releases the descriptors it opened.
~GbmDevice();
};
// ---- format description ------------------------------------------------
// Fills planes[]/planeCount/totalBytes with the linear layout this library
// would allocate for format at width x height, and returns false for a format
// it cannot lay out. gbm_bo_create refuses a format this refuses, because a
// buffer whose plane layout was guessed is worse than a buffer that was not
// handed out at all.
Bool GbmDescribeFormat(Uint32 format, Uint32 width, Uint32 height, GbmPlaneLayout planes[GbmMaxPlanes],
Int& planeCount, SizeT& totalBytes) noexcept;
// Whether this library can allocate the format at all.
Bool GbmFormatIsSupported(Uint32 format) noexcept;
// Bytes per pixel of plane 0: what gbm_bo_map needs to place a sub-rectangle,
// and half of what gbm_bo_get_bpp reports.
Uint32 GbmFormatBytesPerPixel(Uint32 format) noexcept;
// ---- records and the self-allocated backend ----------------------------
// Builds a device. The only helper here that is called without the mutex,
// because no other thread can see the device it is building.
::gbm_device* GbmDeviceCreate(Int fd) noexcept;
// Tears the device down: every remaining surface, every remaining buffer, the
// heap node, the private device fd, and the record itself.
void GbmDeviceDestroy(::gbm_device& device) noexcept;
// Allocates a buffer from the device's heap node, or from a memfd when there
// is none, and records it in the device's table. Returns null with errno set.
::gbm_bo* GbmBufferCreate(::gbm_device& device, Uint32 width, Uint32 height, Uint32 format, Uint64 usage,
const Uint64* modifiers, Uint32 modifierCount) noexcept;
// Wraps descriptors the caller already owns. The descriptors are duped, so the
// caller keeps ownership of its own; type is GBM_BO_IMPORT_FD or
// GBM_BO_IMPORT_FD_MODIFIER.
::gbm_bo* GbmBufferImport(::gbm_device& device, Uint32 type, const void* buffer, Uint64 usage) noexcept;
// Drops one reference, destroying the buffer when it was the last one.
void GbmBufferRelease(::gbm_device& device, ::gbm_bo& bo) noexcept;
// Destroys a buffer outright, whatever its reference count says. Only the
// device's own teardown uses this: by then no caller's reference is meaningful.
void GbmBufferForceDestroy(::gbm_device& device, ::gbm_bo& bo) noexcept;
// Hands out a dup of the descriptor backing a plane, which the caller then owns.
// A plane this buffer does not have is refused with errno = EINVAL.
Int GbmBufferDupFd(::gbm_bo& bo, Int plane) noexcept;
// Maps the whole buffer once and counts the call; strideOut receives the
// stride of plane 0. The rect (x, y) is applied to the returned address.
void* GbmBufferMap(::gbm_bo& bo, Uint32 x, Uint32 y, Uint32& strideOut) noexcept;
// Drops one mapping; the last one unmaps.
void GbmBufferUnmap(::gbm_bo& bo, void* mapData) noexcept;
// ---- host frames -------------------------------------------------------
// Builds the channel to the display host and makes one attempt to connect it, then
// stores it in the device. A machine that is not the container half of a split run
// has nothing at that path, and that is a device which allocates locally from then
// on - not a device that failed to be created.
void GbmHostChannelCreate(::gbm_device& device) noexcept;
// Closes the channel and releases what it holds. Called from the device's
// destructor, after the buffers that still name offered frames are gone, so that
// their releases can still be sent.
void GbmHostChannelDestroy(::gbm_device& device) noexcept;
// Takes the next frame the host offered, or returns false when none is waiting or
// the channel is down. Never blocks: the socket is non-blocking, so a caller that
// finds nothing allocates locally in the same call, which is what makes this usable
// from a compositor's frame loop.
Bool GbmHostChannelTakeFrame(::gbm_device& device, GbmHostFrame& frameOut) noexcept;
// Tells the host that a frame may be recycled. Called from the buffer record's
// destructor, so a frame goes back exactly when nothing in this process can still
// read it, and only into the connection that offered it.
void GbmHostChannelReleaseFrame(::gbm_device& device, Uint32 seq, Uint64 generation) noexcept;
// ---- swapchains --------------------------------------------------------
// Builds a surface and its swapchain up front, so that
// gbm_surface_has_free_buffers can answer truthfully from the first call.
::gbm_surface* GbmSurfaceCreate(::gbm_device& device, Uint32 width, Uint32 height, Uint32 format, Uint64 usage,
const Uint64* modifiers, Uint32 modifierCount) noexcept;
// Unregisters the surface and destroys it, returning the swapchain buffers it
// still owns to the device.
void GbmSurfaceDestroy(::gbm_surface& surface) noexcept;
// Hands out the next free swapchain buffer with a reference of its own, or
// null with errno = EBUSY when the compositor still holds every one of them.
::gbm_bo* GbmSurfaceLockFront(::gbm_surface& surface) noexcept;
// Takes a locked buffer back, dropping the reference the lock handed out.
void GbmSurfaceReleaseBuffer(::gbm_surface& surface, ::gbm_bo& bo) noexcept;
Bool GbmSurfaceHasFreeBuffers(::gbm_surface& surface) noexcept;
} // namespace MobileGL::MG_Gbm
// The public opaque types are the internal records. Callers only ever see the
// pointers, so this costs nothing and removes a layer of casting.
struct gbm_device
{
MobileGL::MG_Gbm::GbmDevice Impl;
};
struct gbm_bo
{
MobileGL::MG_Gbm::GbmBufferObject Impl;
};
struct gbm_surface
{
MobileGL::MG_Gbm::GbmSurface Impl;
};
// End of Source File Header
+408 -13
View File
@@ -7,14 +7,51 @@
// End of Source File Header
#include "EGLImpl.h"
#include "EGLPlatformExtensions.h"
#include "WaylandWindow.h"
#include "../GetProcAddress.h"
#include <Init.h>
#include <MG_Backend/BackendObjects.h>
#include <MG_State/EGLState/Core.h>
#if MOBILEGL_BUILD_DISAGGREGATED
// MG_Config::ServerOwnedWindowSurfaces (P12, MOBILEGL_IPC_SURFACE). Split-only: the pull build's
// translation unit is unchanged (G1).
#include <Config.h>
#endif
#include <mutex>
#include <sstream>
#include <type_traits>
// The KHR spellings of the image entry points, exported by Exporting/Definitions.cpp. Declared
// here because the bundled EGL headers declare them only under EGL_EGLEXT_PROTOTYPES, which no
// build of this library defines, and because eglGetProcAddress below answers them by name.
extern "C" EGLImageKHR eglCreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target, EGLClientBuffer buffer,
const EGLint* attrib_list);
extern "C" EGLBoolean eglDestroyImageKHR(EGLDisplay dpy, EGLImageKHR image);
// The swap-and-sync family the compositor resolves by name while bringing its EGL backend up; defined
// in Exporting/Definitions.cpp and answered from GetProcAddress below (see the definitions).
extern "C" EGLBoolean eglSwapBuffersWithDamageKHR(EGLDisplay dpy, EGLSurface surface, const EGLint* rects,
EGLint n_rects);
extern "C" EGLBoolean eglSwapBuffersWithDamageEXT(EGLDisplay dpy, EGLSurface surface, const EGLint* rects,
EGLint n_rects);
extern "C" EGLSyncKHR eglCreateSyncKHR(EGLDisplay dpy, EGLenum type, const EGLint* attrib_list);
extern "C" EGLBoolean eglDestroySyncKHR(EGLDisplay dpy, EGLSyncKHR sync);
extern "C" EGLint eglClientWaitSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint flags, EGLTimeKHR timeout);
extern "C" EGLBoolean eglWaitSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint flags);
extern "C" EGLBoolean eglGetSyncAttribKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint attribute, EGLint* value);
// THE DEVICE-QUERY FAMILY, exported by Exporting/Definitions.cpp and answered by name below: kwin
// reaches eglQueryDisplayAttribEXT through libepoxy's pointer (that is how finding a render device
// starts), and libepoxy asks for it by name. See the definitions for why these had to exist.
extern "C" EGLBoolean eglQueryDevicesEXT(EGLint max_devices, EGLDeviceEXT* devices, EGLint* num_devices);
extern "C" EGLBoolean eglQueryDeviceAttribEXT(EGLDeviceEXT device, EGLint attribute, EGLAttrib* value);
extern "C" const char* eglQueryDeviceStringEXT(EGLDeviceEXT device, EGLint name);
extern "C" EGLBoolean eglQueryDisplayAttribEXT(EGLDisplay dpy, EGLint attribute, EGLAttrib* value);
extern "C" EGLSurface eglCreatePlatformWindowSurfaceEXT(EGLDisplay dpy, EGLConfig config, void* native_window,
const EGLAttrib* attrib_list);
extern "C" EGLSurface eglCreatePlatformPixmapSurfaceEXT(EGLDisplay dpy, EGLConfig config, void* native_pixmap,
const EGLAttrib* attrib_list);
namespace MobileGL::MG_Impl::EGLImpl {
namespace {
using EGLStateContext = MG_State::EGLState::EGLContext;
@@ -111,6 +148,73 @@ namespace MobileGL::MG_Impl::EGLImpl {
return reinterpret_cast<void*>(static_cast<SizeT>(nativeHandle));
}
}
#if MOBILEGL_WAYLAND_WINDOWS
// Every Wayland window surface's presentation, by surface. Guarded by EGLOperationMutex,
// which every entry point that reaches it holds.
UnorderedMap<EGLSurface, UniquePtr<Wayland::WindowSurface>>& WaylandSurfaces() {
static auto* surfaces = new UnorderedMap<EGLSurface, UniquePtr<Wayland::WindowSurface>>();
return *surfaces;
}
// A WINDOW ON A WAYLAND DISPLAY. Whatever MOBILEGL_IPC_SURFACE says, the window is this
// application's own wl_surface, so neither the server's window nor the display host's frame
// is what it shows: the backend draws into a pbuffer of the window's size, and each swap
// reads the frame back and attaches it to the wl_surface (WaylandWindow.h). Answers false
// when the display is not a Wayland one; otherwise *out is the surface, or EGL_NO_SURFACE
// with the error set.
template <typename State>
Bool CreateWaylandWindowSurface(State* state, EGLDisplay dpy, EGLConfig config, void* window,
Bool platformWindow, EGLSurface* out) {
Uint64 nativeDisplay = 0;
EGLenum platform = EGL_NONE;
if (!state->GetDisplayNative(dpy, &nativeDisplay, &platform) ||
!Wayland::IsWaylandDisplay(nativeDisplay, platform)) {
return false;
}
*out = EGL_NO_SURFACE;
if (!state->IsDisplayInitialized(dpy)) {
state->SetError(EGL_NOT_INITIALIZED);
return true;
}
if (!state->ValidateConfigOnDisplay(dpy, config)) {
state->SetError(EGL_BAD_CONFIG);
return true;
}
EGLint width = 0;
EGLint height = 0;
if (!Wayland::WindowSize(window, &width, &height)) {
state->SetError(EGL_BAD_NATIVE_WINDOW);
return true;
}
EGLint alpha = 0;
(void)state->GetConfigAttrib(dpy, config, EGL_ALPHA_SIZE, &alpha);
auto presentation =
Wayland::WindowSurface::Create(reinterpret_cast<void*>(static_cast<uintptr_t>(nativeDisplay)), window,
width, height, alpha > 0);
if (!presentation) {
state->SetError(EGL_BAD_NATIVE_WINDOW);
return true;
}
const EGLSurface surface =
platformWindow ? state->CreatePlatformWindowSurface(dpy, config, window, nullptr)
: state->CreateWindowSurface(
dpy, config, (NativeWindowType)(reinterpret_cast<uintptr_t>(window)), nullptr);
if (surface == EGL_NO_SURFACE) return true;
(void)state->ResizeSurface(dpy, surface, width, height);
auto* backendObject = GetBackendObject(state);
if (!backendObject || !backendObject->CreateEGLPbufferSurface(surface, width, height)) {
MGLOG_E("eglCreateWindowSurface: the backend refused the %dx%d drawable behind a Wayland window",
width, height);
state->DestroySurface(dpy, surface);
state->SetError(EGL_BAD_ALLOC);
return true;
}
WaylandSurfaces()[surface] = std::move(presentation);
*out = surface;
return true;
}
#endif
} // namespace
EGLSurface CreateWindowSurface(EGLDisplay dpy, EGLConfig config, NativeWindowType window,
@@ -119,6 +223,16 @@ namespace MobileGL::MG_Impl::EGLImpl {
if (!state) {
return EGL_NO_SURFACE;
}
#if MOBILEGL_WAYLAND_WINDOWS
{
const std::lock_guard<std::recursive_mutex> operationLock(EGLOperationMutex());
EGLSurface waylandSurface = EGL_NO_SURFACE;
if (CreateWaylandWindowSurface(state, dpy, config, ToVoidHandle(window), /*platformWindow=*/false,
&waylandSurface)) {
return waylandSurface;
}
}
#endif
if (!state->IsDisplayInitialized(dpy)) {
state->SetError(EGL_NOT_INITIALIZED);
return EGL_NO_SURFACE;
@@ -127,19 +241,71 @@ namespace MobileGL::MG_Impl::EGLImpl {
state->SetError(EGL_BAD_CONFIG);
return EGL_NO_SURFACE;
}
#if MOBILEGL_BUILD_DISAGGREGATED
// P12 (on-screen server window), D1. With MOBILEGL_IPC_SURFACE=server and a remote server,
// the window is the SERVER's: a headless client passes none, so NULL is not
// EGL_BAD_NATIVE_WINDOW here. Whatever the client passed never reaches the wire - the
// remote backend sends WindowKind::ServerOwned with token 0 (BackendObject_Remote).
// Either on-screen shape leaves this side without a window of its own: =server asks the
// server for its own window, =host draws into the frames the display host offers.
const Bool windowlessOnScreen =
MG_Config::ServerOwnedWindowSurfaces() || MG_Config::HostFrameWindowSurfaces();
if (IsNullNativeHandle(window) && !windowlessOnScreen) {
#else
if (IsNullNativeHandle(window)) {
#endif
state->SetError(EGL_BAD_NATIVE_WINDOW);
return EGL_NO_SURFACE;
}
Uint32 requestedWidth = static_cast<Uint32>(std::max<EGLint>(GetAttribValue(attrib_list, EGL_WIDTH, 0), 0));
Uint32 requestedHeight = static_cast<Uint32>(std::max<EGLint>(GetAttribValue(attrib_list, EGL_HEIGHT, 0), 0));
#if MOBILEGL_BUILD_DISAGGREGATED && defined(__ANDROID__)
// P12: A GAME'S OWN WINDOW SIZE IS THE SIZE OF THE FRAMES IT RENDERS, and a server-owned
// window has to take that size or clip. SDL (Minecraft's Android backend) creates the EGL
// window surface with no EGL_WIDTH / EGL_HEIGHT, so the request went out as 0x0 - "use the
// server window's own size" - and a landscape 2620x1280 game drawn into a portrait 1280x2620
// server window came out clipped to its bottom-left 1280x1280 corner (Espryt and Magma
// alike; the same game in a landscape window was right). The native window it handed us
// knows the size the game believes it has: ask for THAT, and the server fixes its window
// buffer to it and aspect-fits the view (MobileGLDisplayActivity), letterboxed.
if (windowlessOnScreen && requestedWidth == 0 && requestedHeight == 0 && !IsNullNativeHandle(window)) {
auto* native = reinterpret_cast<ANativeWindow*>(ToVoidHandle(window));
const int32_t nativeWidth = ANativeWindow_getWidth(native);
const int32_t nativeHeight = ANativeWindow_getHeight(native);
if (nativeWidth > 0 && nativeHeight > 0) {
requestedWidth = static_cast<Uint32>(nativeWidth);
requestedHeight = static_cast<Uint32>(nativeHeight);
MGLOG_I("eglCreateWindowSurface: no EGL_WIDTH/EGL_HEIGHT; the on-screen surface is asked for "
"the native window's %dx%d", nativeWidth, nativeHeight);
}
}
#endif
const MG_Backend::WindowHandle windowHandle = {
.Backend = DetectWindowBackend(),
.Handle = ToVoidHandle(window),
.Width = static_cast<Uint32>(std::max<EGLint>(GetAttribValue(attrib_list, EGL_WIDTH, 0), 0)),
.Height = static_cast<Uint32>(std::max<EGLint>(GetAttribValue(attrib_list, EGL_HEIGHT, 0), 0)),
.Width = requestedWidth,
.Height = requestedHeight,
};
#if MOBILEGL_BUILD_DISAGGREGATED
EGLSurface surface = EGL_NO_SURFACE;
if (MG_Config::ServerOwnedWindowSurfaces()) {
// The server's own window: it makes one and draws into it.
surface = state->CreateServerOwnedWindowSurface(
dpy, config, windowHandle.Handle, static_cast<EGLint>(windowHandle.Width),
static_cast<EGLint>(windowHandle.Height), /*platformWindow=*/false);
} else if (MG_Config::HostFrameWindowSurfaces()) {
// The display host's frames, as above: no window of the client's either way.
surface = state->CreateHostFrameWindowSurface(
dpy, config, windowHandle.Handle, static_cast<EGLint>(windowHandle.Width),
static_cast<EGLint>(windowHandle.Height), /*platformWindow=*/false);
} else {
surface = state->CreateWindowSurface(dpy, config, window, attrib_list);
}
#else
EGLSurface surface = state->CreateWindowSurface(dpy, config, window, attrib_list);
#endif
if (surface == EGL_NO_SURFACE) {
return EGL_NO_SURFACE;
}
@@ -166,6 +332,7 @@ namespace MobileGL::MG_Impl::EGLImpl {
return EGL_FALSE;
}
if (!state->ValidateSurfaceOnDisplay(dpy, draw)) {
MGLOG_E_ONCE("eglSwapBuffers: surface %p is not a surface of display %p (EGL_BAD_SURFACE)", draw, dpy);
state->SetError(EGL_BAD_SURFACE);
return EGL_FALSE;
}
@@ -175,6 +342,11 @@ namespace MobileGL::MG_Impl::EGLImpl {
MGLOG_E_ONCE("activeBackendObject not initialized!");
return EGL_FALSE;
}
#if MOBILEGL_WAYLAND_WINDOWS
// A Wayland window shows what is attached to it: the frame is read back and attached
// BEFORE the swap, while it is still the drawable's content.
if (auto it = WaylandSurfaces().find(draw); it != WaylandSurfaces().end()) (void)it->second->Present();
#endif
if (!backendObject->SwapEGLBuffers(dpy, draw)) {
MGLOG_E_ONCE("eglSwapBuffers failed on thread=%s dpy=%p draw=%p", CurrentThreadIdString().c_str(), dpy, draw);
state->SetError(EGL_BAD_SURFACE);
@@ -185,7 +357,13 @@ namespace MobileGL::MG_Impl::EGLImpl {
EGLBoolean ChooseConfig(EGLDisplay dpy, const EGLint* attrib_list, EGLConfig* configs, EGLint config_size,
EGLint* num_config) {
auto* state = GetState();
// eglChooseConfig can be an application's first EGL call after eglGetPlatformDisplay + eglInitialize,
// and for a display built from an EGLDeviceEXT nothing has brought MobileGL up yet: GetState()
// then answers NULL and the call lands in EGLContext::ChooseConfig with a null this - measured as
// a crash inside that function (pc/lr both resolved to it) with kwin's own "pEGLContext is null.
// MG_State may not be initialized." as the only warning, and reported one layer up as
// "Could not initialize rendering context".
auto* state = GetStateEnsureInitialized();
if (!state) {
return EGL_FALSE;
}
@@ -193,11 +371,26 @@ namespace MobileGL::MG_Impl::EGLImpl {
}
EGLContext CreateContext(EGLDisplay dpy, EGLConfig config, EGLContext shareCtx, const EGLint* attrib_list) {
auto* state = GetState();
// WHICH CONTEXT IS BEING ASKED FOR, and what came back. kwin's OpenGL compositing fails with
// "Could not initialize rendering context" while the same path succeeds from a probe against
// this same library (probe10: a 3.3 core context on a GBM surface, makeCurrent answered 1),
// so the difference has to be in this argument list - and an EGL vendor cannot see it from the
// outside. The list is dumped in EGL_NONE-terminated pairs; EGL_NONE is 0x3038, not 0.
MGLOG_I("eglCreateContext(dpy=%p, config=%p, share=%p)", dpy, config, shareCtx);
if (attrib_list != nullptr) {
for (int index = 0; attrib_list[index] != EGL_NONE; index += 2) {
MGLOG_I(" attrib 0x%04x = %d", attrib_list[index], attrib_list[index + 1]);
}
} else {
MGLOG_I(" (no attribute list)");
}
auto* state = GetStateEnsureInitialized();
if (!state) {
return EGL_NO_CONTEXT;
}
return state->CreateContext(dpy, config, shareCtx, attrib_list);
EGLContext context = state->CreateContext(dpy, config, shareCtx, attrib_list);
MGLOG_I("eglCreateContext -> %p", static_cast<void*>(context));
return context;
}
EGLBoolean Initialize(EGLDisplay dpy, EGLint* major, EGLint* minor) {
@@ -237,6 +430,13 @@ namespace MobileGL::MG_Impl::EGLImpl {
return state->ConsumeError();
}
// THE DISPLAY HOST'S FRAME, AS THE COMPOSITOR PRESENTS IT. Set by the surfaceless-bind hook in
// MakeCurrent when this client (the compositor) takes the host frame; read by glFlush, which is
// that compositor's only frame boundary - it never calls eglSwapBuffers, and without a present
// the display host is never told a frame is done, so nothing ever reaches the glass.
EGLDisplay g_hostFramePresentDisplay = nullptr;
EGLSurface g_hostFramePresentSurface = nullptr;
EGLBoolean MakeCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read, EGLContext ctx) {
const std::lock_guard<std::recursive_mutex> operationLock(EGLOperationMutex());
auto* state = GetState();
@@ -253,6 +453,81 @@ namespace MobileGL::MG_Impl::EGLImpl {
MGLOG_D("eglMakeCurrent begin thread=%s dpy=%p draw=%p read=%p ctx=%p oldDpy=%p oldDraw=%p oldRead=%p oldCtx=%p",
threadId.c_str(), dpy, draw, read, ctx, oldDisplay, oldDraw, oldRead, oldContext);
#if MOBILEGL_BUILD_DISAGGREGATED
// THE DISPLAY HOST'S FRAMES, TO A CLIENT THAT ASKS FOR NO SURFACE.
//
// A compositor on this route never calls eglCreateWindowSurface: it imports the display host's
// dma-bufs and draws into them through framebuffers, binding its contexts with EGL_NO_SURFACE.
// The host frame bridge, though, is opened by exactly one thing - a surface being CREATED and
// activated (ServerLoop's CreateWindowSurface with the host-frame window backend ->
// InitHostFrameSurface) - so with no surface ever created the bridge stayed shut, the display
// host kept offering frames into a name nobody served ("offer buffer 0: bridge fd=-1"), and no
// frame could reach the glass no matter how much was drawn. So: when the session asked for
// host-framed surfaces and a real context is being bound with no surface, make it the surface
// eglCreatePlatformWindowSurface would have made, once, and bind that. The size is the display
// host's own frame size; the client's window value is not on this path at all.
// ONLY THE COMPOSITOR TAKES THE HOST FRAME. A desktop is full of processes that bind a
// context with no surface - konsole, plasmashell, kactivitymanagerd - and measured with
// probe15 as the second client: such a bind from a NON-compositor process made the host-frame
// activation fail (makeCurrent=0) and the process aborted. The compositor is marked by its
// launcher with MOBILEGL_HOST_FRAME_CLIENT=1; everyone else keeps the plain surfaceless bind,
// which is all they ever wanted.
if (draw == EGL_NO_SURFACE && read == EGL_NO_SURFACE && ctx != EGL_NO_CONTEXT &&
MG_Config::HostFrameWindowSurfaces() &&
(std::getenv("MOBILEGL_HOST_FRAME_CLIENT") != nullptr &&
std::string("1") == std::getenv("MOBILEGL_HOST_FRAME_CLIENT"))) {
static Bool hostFrameSurfaceTaken = false;
if (!hostFrameSurfaceTaken) {
hostFrameSurfaceTaken = true;
EGLConfig hostFrameConfig = nullptr;
EGLint hostFrameConfigCount = 0;
const EGLint hostFrameChoose[] = {EGL_SURFACE_TYPE, EGL_WINDOW_BIT, EGL_NONE};
if (state->ChooseConfig(dpy, hostFrameChoose, &hostFrameConfig, 1,
&hostFrameConfigCount) &&
hostFrameConfigCount > 0 && hostFrameConfig != nullptr) {
// BOTH HALVES, exactly as eglCreateWindowSurface makes one: the EGL state's
// registration (without it every later bind and swap of this surface was refused
// as EGL_BAD_SURFACE - measured: "eglMakeCurrent rejected by EGLState
// error=0x300d", and glFlush's swaps returned before reaching the backend), then
// the backend's own window-surface entry, which the remote backend routes to the
// host frame (BackendObject_Remote::CreateHostFrameWindowSurface) - the call that
// puts a CreateWindowSurface frame with the host-frame window backend on the wire.
const EGLSurface hostFrameSurface = state->CreateHostFrameWindowSurface(
dpy, hostFrameConfig, nullptr, 1440, 3200, /*platformWindow=*/false);
MG_Backend::WindowHandle hostFrameHandle{};
hostFrameHandle.Width = 1440;
hostFrameHandle.Height = 3200;
if (hostFrameSurface == EGL_NO_SURFACE) {
MGLOG_E("host frame: the EGL state refused the display host's frame surface");
} else if (!MG_Backend::pActiveBackendObject->CreateEGLWindowSurface(hostFrameSurface,
hostFrameHandle)) {
MGLOG_E("host frame: the backend refused the display host's frame surface");
state->DestroySurface(dpy, hostFrameSurface);
} else {
// The frame boundary of a compositor that never swaps is its glFlush. glFlush
// has no EGL context of its own, so remember HERE which surface presents the
// display host's frame; glFlush presents it (see the host-frame present in
// glFlush's implementation).
g_hostFramePresentDisplay = dpy;
g_hostFramePresentSurface = hostFrameSurface;
MGLOG_I("host frame: the first surfaceless eglMakeCurrent made the display host's "
"frame surface %p (%dx%d)",
hostFrameSurface, 1440, 3200);
}
} else {
MGLOG_E("host frame: no config for the display host's frame surface");
}
}
// EVERY surfaceless bind, not only the first: the compositor rebinds with EGL_NO_SURFACE
// all the time, and a bind without the surface leaves nothing current for glFlush's swap
// to present (BackendObject::SwapEGLBuffers: "draw surface is not current on this thread").
if (g_hostFramePresentSurface != nullptr && dpy == g_hostFramePresentDisplay) {
draw = g_hostFramePresentSurface;
read = g_hostFramePresentSurface;
}
}
#endif
if (!state->MakeCurrent(dpy, draw, read, ctx)) {
const EGLint error = state->ConsumeError();
MGLOG_D("eglMakeCurrent rejected by EGLState thread=%s error=0x%04x", threadId.c_str(), error);
@@ -310,6 +585,9 @@ namespace MobileGL::MG_Impl::EGLImpl {
if (!state->DestroySurface(dpy, surface)) {
return EGL_FALSE;
}
#if MOBILEGL_WAYLAND_WINDOWS
WaylandSurfaces().erase(surface);
#endif
if (auto* backendObject = MG_Backend::pActiveBackendObject.get()) {
backendObject->ReleaseEGLSurface(surface);
}
@@ -418,14 +696,10 @@ namespace MobileGL::MG_Impl::EGLImpl {
case EGL_CLIENT_APIS:
return "OpenGL OpenGL_ES";
case EGL_EXTENSIONS:
if (display == EGL_NO_DISPLAY) {
return "EGL_EXT_client_extensions "
"EGL_EXT_platform_base "
"EGL_KHR_platform_base "
"EGL_MESA_platform_surfaceless";
}
return "EGL_KHR_create_context "
"EGL_MESA_platform_surfaceless";
// One list, in EGLPlatformExtensions.h, because the vendor string glvnd reads and
// this query an application reads have to agree: an application that does not see the
// platform extension here never asks for a platform display at all.
return display == EGL_NO_DISPLAY ? kClientExtensionString : kDisplayExtensionString;
default:
state->SetError(EGL_BAD_PARAMETER);
return nullptr;
@@ -660,6 +934,23 @@ namespace MobileGL::MG_Impl::EGLImpl {
return state->DestroyImage(dpy, image) ? EGL_TRUE : EGL_FALSE;
}
EGLImageKHR CreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target, EGLClientBuffer buffer,
const EGLint* attrib_list) {
// EGL_KHR_image_base spells its attribute list in EGLint pairs and the core eglCreateImage
// in EGLAttrib pairs, and the two are not the same memory: on a 64-bit build one entry is
// twice the width of the other, so the caller's array cannot be handed over as it stands.
// Re-spelling it is therefore part of the entry point rather than a detail of the state.
Vector<EGLAttrib> attribs;
if (attrib_list != nullptr) {
for (SizeT i = 0; attrib_list[i] != EGL_NONE; i += 2) {
attribs.push_back(static_cast<EGLAttrib>(attrib_list[i]));
attribs.push_back(static_cast<EGLAttrib>(attrib_list[i + 1]));
}
attribs.push_back(static_cast<EGLAttrib>(EGL_NONE));
}
return CreateImage(dpy, ctx, target, buffer, attribs.empty() ? nullptr : attribs.data());
}
EGLDisplay GetPlatformDisplay(EGLenum platform, void* native_display, const EGLAttrib* attrib_list) {
(void)attrib_list;
@@ -676,7 +967,27 @@ namespace MobileGL::MG_Impl::EGLImpl {
if (!state) {
return EGL_NO_SURFACE;
}
#if MOBILEGL_WAYLAND_WINDOWS
{
const std::lock_guard<std::recursive_mutex> operationLock(EGLOperationMutex());
EGLSurface waylandSurface = EGL_NO_SURFACE;
if (CreateWaylandWindowSurface(state, dpy, config, native_window, /*platformWindow=*/true,
&waylandSurface)) {
return waylandSurface;
}
}
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// P12 (on-screen server window), D1: as CreateWindowSurface - with MOBILEGL_IPC_SURFACE=server
// the window is the server's, and a headless client's NULL is accepted.
// Either on-screen shape leaves this side without a window of its own: =server asks the
// server for its own window, =host draws into the frames the display host offers.
const Bool windowlessOnScreen =
MG_Config::ServerOwnedWindowSurfaces() || MG_Config::HostFrameWindowSurfaces();
if (native_window == nullptr && !windowlessOnScreen) {
#else
if (native_window == nullptr) {
#endif
state->SetError(EGL_BAD_NATIVE_WINDOW);
return EGL_NO_SURFACE;
}
@@ -696,7 +1007,25 @@ namespace MobileGL::MG_Impl::EGLImpl {
.Height = static_cast<Uint32>(std::max<EGLint>(GetAttribValueAttrib(attrib_list, EGL_HEIGHT, 0), 0)),
};
#if MOBILEGL_BUILD_DISAGGREGATED
EGLSurface surface = EGL_NO_SURFACE;
if (MG_Config::ServerOwnedWindowSurfaces()) {
// The server's own window: it makes one and draws into it.
surface = state->CreateServerOwnedWindowSurface(
dpy, config, native_window, static_cast<EGLint>(windowHandle.Width),
static_cast<EGLint>(windowHandle.Height), /*platformWindow=*/true);
} else if (MG_Config::HostFrameWindowSurfaces()) {
// The display host's frames: it allocates them, the server draws into them,
// and the host puts each one on the glass when the server says it is drawn.
surface = state->CreateHostFrameWindowSurface(
dpy, config, native_window, static_cast<EGLint>(windowHandle.Width),
static_cast<EGLint>(windowHandle.Height), /*platformWindow=*/true);
} else {
surface = state->CreatePlatformWindowSurface(dpy, config, native_window, attrib_list);
}
#else
EGLSurface surface = state->CreatePlatformWindowSurface(dpy, config, native_window, attrib_list);
#endif
if (surface == EGL_NO_SURFACE) {
return EGL_NO_SURFACE;
}
@@ -762,6 +1091,72 @@ namespace MobileGL::MG_Impl::EGLImpl {
MobileGL::EnsureInitialized();
MGLOG_D("eglGetProcAddress(%s)", name);
// THE KHR SPELLINGS ARE ENTRY POINTS OF THEIR OWN, and by name is the only way anything
// asks for them. Measured in the Fedora container that runs this library as the glvnd EGL
// vendor, before they were answered here: eglGetProcAddress("eglCreateImageKHR") -> NULL and
// eglGetProcAddress("eglDestroyImageKHR") -> NULL. kwin_wayland's anland backend imports
// the display daemon's dma-bufs through exactly these two (AnlandEglLayer::importBuffers ->
// EglBackend::importDmaBufAsTexture -> EglDisplay::importBufferAsImage ->
// EglDisplay::createImage, which calls the pointer without a null check), so a NULL here is
// a compositor that cannot draw a single pixel into the frames the daemon hands it. The
// core spellings were already in the table below; these are the same functions under the
// names EGL_KHR_image_base gives them, which is also why they answer in the glvnd vendor
// table beside them (Exporting/GlvndVendor.cpp).
if (std::strcmp(name, "eglCreateImageKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglCreateImageKHR);
}
if (std::strcmp(name, "eglDestroyImageKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglDestroyImageKHR);
}
// The device-query family and the EXT platform-surface spellings. Measured in the container
// before they were answered here: eglGetProcAddress returned NULL for eglQueryDisplayAttribEXT,
// eglQueryDevicesEXT, eglQueryDeviceAttribEXT, eglQueryDeviceStringEXT and
// eglCreatePlatformWindowSurfaceEXT while the extension strings announced EGL_EXT_device_query -
// a NULL libepoxy then called (kwin crashed at pc=0), and withdrawing the announcement instead
// left kwin without a render device and its compositing inactive.
if (std::strcmp(name, "eglQueryDevicesEXT") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglQueryDevicesEXT);
}
if (std::strcmp(name, "eglQueryDeviceAttribEXT") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglQueryDeviceAttribEXT);
}
if (std::strcmp(name, "eglQueryDeviceStringEXT") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglQueryDeviceStringEXT);
}
if (std::strcmp(name, "eglQueryDisplayAttribEXT") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglQueryDisplayAttribEXT);
}
if (std::strcmp(name, "eglCreatePlatformWindowSurfaceEXT") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglCreatePlatformWindowSurfaceEXT);
}
if (std::strcmp(name, "eglCreatePlatformPixmapSurfaceEXT") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglCreatePlatformPixmapSurfaceEXT);
}
// The swap-and-sync spellings kwin resolves by name while it is still bringing its EGL backend
// up (see the definitions: EGL_KHR_fence_sync, the damage variants, EGL_EXT_buffer_age).
if (std::strcmp(name, "eglSwapBuffersWithDamageKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglSwapBuffersWithDamageKHR);
}
if (std::strcmp(name, "eglSwapBuffersWithDamageEXT") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglSwapBuffersWithDamageEXT);
}
if (std::strcmp(name, "eglCreateSyncKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglCreateSyncKHR);
}
if (std::strcmp(name, "eglDestroySyncKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglDestroySyncKHR);
}
if (std::strcmp(name, "eglClientWaitSyncKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglClientWaitSyncKHR);
}
if (std::strcmp(name, "eglWaitSyncKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglWaitSyncKHR);
}
if (std::strcmp(name, "eglGetSyncAttribKHR") == 0) {
return reinterpret_cast<__eglMustCastToProperFunctionPointerType>(eglGetSyncAttribKHR);
}
void* proc = MG_Impl::GetProcAddress(name);
if (!proc) {
MGLOG_D("Failed to get function: %s", name);
+14
View File
@@ -8,8 +8,18 @@
#pragma once
#include <Includes.h>
// EGLImageKHR: the return type of the KHR spelling of eglCreateImage below. Includes.h brings in
// the core egl.h, which has no KHR types in it.
#include <EGL/eglext.h>
namespace MobileGL::MG_Impl::EGLImpl {
// The display host's frame surface, as taken by the surfaceless-bind hook in MakeCurrent, and the
// display it belongs to. glFlush reads these: a compositor's frame boundary is glFlush, and the
// host frame's "frame is done" signal IS a swap of this surface. Null until a client marked
// MOBILEGL_HOST_FRAME_CLIENT=1 actually took the frame.
extern EGLDisplay g_hostFramePresentDisplay;
extern EGLSurface g_hostFramePresentSurface;
EGLSurface CreateWindowSurface(EGLDisplay dpy, EGLConfig config, NativeWindowType window,
const EGLint* attrib_list);
EGLBoolean SwapBuffers(EGLDisplay dpy, EGLSurface draw);
@@ -53,6 +63,10 @@ namespace MobileGL::MG_Impl::EGLImpl {
EGLBoolean GetSyncAttrib(EGLDisplay dpy, EGLSync sync, EGLint attribute, EGLAttrib* value);
EGLImage CreateImage(EGLDisplay dpy, EGLContext ctx, EGLenum target, EGLClientBuffer buffer,
const EGLAttrib* attrib_list);
// The same image under EGL_KHR_image_base's name and its EGLint attribute list: the two lists
// are pairs of different widths, so the caller's array is re-spelled rather than cast.
EGLImageKHR CreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target, EGLClientBuffer buffer,
const EGLint* attrib_list);
EGLBoolean DestroyImage(EGLDisplay dpy, EGLImage image);
EGLDisplay GetPlatformDisplay(EGLenum platform, void* native_display, const EGLAttrib* attrib_list);
EGLSurface CreatePlatformWindowSurface(EGLDisplay dpy, EGLConfig config, void* native_window,
@@ -0,0 +1,158 @@
// MobileGL - MobileGL/MG_Impl/EGLImpl/EGLPlatformExtensions.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
namespace MobileGL::MG_Impl::EGLImpl {
// THE PLATFORMS THIS IMPLEMENTATION SERVES, in the words EGL spells them.
//
// Two callers need exactly the same list, and they disagree if it is written twice:
// eglQueryString(EGL_EXTENSIONS) is what an application reads BEFORE it decides to call
// eglGetPlatformDisplay at all - libepoxy refuses to use eglGetPlatformDisplayEXT unless the
// string names EGL_EXT_platform_base - and the glvnd vendor string is what the system's
// libEGL reads to decide which vendor library a platform request should be routed to. A
// vendor that does not claim the platform is never asked for it, and the caller is told there
// is no provider: the library can be loaded, initialized, and still never called.
//
// GBM is on the list because a compositor asks for its display on the GBM device it owns.
// This side treats that device the way it treats every other native display - the frames come
// from the display host, not from the device - so the claim is exact rather than optimistic.
inline constexpr const char* kDeviceExtensionString =
"EGL_EXT_device_base "
"EGL_EXT_device_enumeration "
"EGL_EXT_device_query "
"EGL_EXT_device_drm "
"EGL_EXT_device_drm_render_node";
// EGL_EXT_client_extensions IS ON THE PLATFORM LIST FOR A MEASURED REASON. libglvnd only folds
// a vendor client-extension string (the answer to eglQueryString(EGL_NO_DISPLAY, EGL_EXTENSIONS))
// into its own no-display answer when the vendor PLATFORM string - the one __egl_Main publishes
// for the platform-extensions query - names EGL_EXT_client_extensions. Without it the client
// string never reached libepoxy, so libepoxy concluded EGL_EXT_device_query was absent and left
// epoxy_eglQueryDisplayAttribEXT as a NULL pointer, while kwin (which reads the display string,
// where the extension IS present) called it: measured as signal 11, fault_addr=(nil), pc=0x0,
// from KWin::EglDisplay::determineRenderNode() through the GOT slot whose relocation is
// R_AARCH64_GLOB_DAT epoxy_eglQueryDisplayAttribEXT.
// Wayland where a Wayland client can exist (WaylandWindow.h: window surfaces presented through
// wl_shm). Without it glvnd never hands this vendor a Wayland display at all.
#if defined(__linux__) && !defined(__ANDROID__)
#define MOBILEGL_PLATFORM_WAYLAND_EXTENSIONS " EGL_KHR_platform_wayland EGL_EXT_platform_wayland"
#else
#define MOBILEGL_PLATFORM_WAYLAND_EXTENSIONS ""
#endif
inline constexpr const char* kPlatformExtensionString =
"EGL_EXT_client_extensions "
"EGL_EXT_platform_base "
"EGL_KHR_platform_base "
"EGL_KHR_platform_gbm "
"EGL_EXT_platform_gbm "
"EGL_MESA_platform_gbm "
"EGL_EXT_platform_device "
"EGL_MESA_platform_surfaceless" MOBILEGL_PLATFORM_WAYLAND_EXTENSIONS;
// What a display answers for EGL_EXTENSIONS: the extensions this implementation provides that
// are not platform ones, then the platform list above.
// EGL_KHR_no_config_context and EGL_KHR_surfaceless_context are on the display and client lists
// because both are answered by this implementation rather than announced by a driver: a context
// is created with EGL_NO_CONFIG_KHR and made current with EGL_NO_SURFACE, which is the pair a
// compositor asks for when it renders into framebuffers of its own. Measured on kwin_wayland
// against this vendor before they were here: createContext() refused with
// "EGL_KHR_no_config_context extension is unsupported" and the session ended at
// "no usable DRM render device; cannot bring up OpenGL compositing".
//
// EGL_KHR_image_base AND EGL_EXT_image_dma_buf_import ARE ONE FEATURE HERE, and they are on the
// list for a measured reason rather than a plausible one. In the Fedora container that runs
// this library as the glvnd EGL vendor, eglGetProcAddress("eglCreateImageKHR") and
// eglGetProcAddress("eglDestroyImageKHR") both returned NULL, and kwin_wayland 6.7.3's anland
// backend imports the display daemon's dma-bufs through exactly those two
// (AnlandEglLayer::importBuffers -> EglBackend::importDmaBufAsTexture ->
// EglDisplay::importBufferAsImage -> EglDisplay::createImage) before it can draw a pixel into a
// frame the daemon produced - so with a NULL the whole composited scene has no image to be
// drawn into. Both names are backed by entry points that exist: EGL_KHR_image_base by
// eglCreateImage/eglDestroyImage and their KHR spellings (Exporting/Definitions.cpp, and the
// same names in the glvnd vendor table), EGL_EXT_image_dma_buf_import by EGL_LINUX_DMA_BUF_EXT
// images, which carry the caller's EGL_WIDTH, EGL_HEIGHT and EGL_LINUX_DRM_FOURCC_EXT and own a
// dup() of every plane's fd, offset and pitch (EGLState/Core.cpp, CreateImage).
//
// EGL_EXT_image_dma_buf_import_modifiers IS DELIBERATELY NOT NAMED. Its own two entry points -
// eglQueryDmaBufFormatsEXT and eglQueryDmaBufModifiersEXT - do not exist in this library, and a
// name in this string means the entry points beside it are there. The modifier attributes
// (EGL_DMA_BUF_PLANEn_MODIFIER_LO/HI_EXT) are nevertheless ACCEPTED and carried on the image:
// a caller does not need the name to send them, and kwin's does not look at any extension
// string before it does - EglDisplay::importDmaBufAsImage appends the pair whenever the
// buffer's modifier is not DRM_FORMAT_MOD_INVALID.
inline constexpr const char* kDisplayExtensionString =
"EGL_KHR_create_context "
// WHAT THE COMPOSITOR CHECKS BETWEEN ITS CONTEXT AND ITS FIRST SURFACE. EglBackend::init
// asks for these three by name right after initRenderingContext() and before createSurfaces();
// the entry points behind them are published beside this string (Exporting/Definitions.cpp),
// so the announcement and the answers are one thing rather than two.
"EGL_KHR_fence_sync "
"EGL_KHR_wait_sync "
"EGL_KHR_swap_buffers_with_damage "
"EGL_EXT_swap_buffers_with_damage "
"EGL_EXT_buffer_age "
"EGL_KHR_image_base "
"EGL_EXT_image_dma_buf_import "
"EGL_KHR_no_config_context "
"EGL_KHR_surfaceless_context "
// ANNOUNCED BECAUSE IT IS NOW ANSWERED: the four names below are published as real entry
// points and eglGetProcAddress answers them (see Exporting/Definitions.cpp and
// EGLImpl::GetProcAddress). Both earlier states were measured failures: announcing the
// extension while the names resolved to NULL made kwin jump to that NULL (signal 11,
// pc=0x0, from KWin::EglDisplay::determineRenderNode()), and withdrawing the announcement
// left kwin without a render device so its compositing never became active and nothing was
// ever painted.
"EGL_EXT_device_base "
"EGL_EXT_device_enumeration "
"EGL_EXT_device_query "
"EGL_EXT_device_drm "
"EGL_EXT_device_drm_render_node "
// KWin::EglDisplay::determineRenderNode() at libkwin.so.6+0x308b9c. A compositor that is
// never told the extension exists cannot take that path; announcing it again is the right
// move once these four names are really published and resolvable.
"EGL_EXT_platform_device "
"EGL_EXT_platform_base "
"EGL_KHR_platform_base "
"EGL_KHR_platform_gbm "
"EGL_EXT_platform_gbm "
"EGL_MESA_platform_gbm "
"EGL_MESA_platform_surfaceless";
// ... and what EGL_NO_DISPLAY answers: the same, plus the client-extension marker that tells a
// caller the no-display query is meaningful here.
inline constexpr const char* kClientExtensionString =
"EGL_EXT_client_extensions "
"EGL_KHR_create_context "
"EGL_KHR_image_base "
"EGL_EXT_image_dma_buf_import "
"EGL_KHR_no_config_context "
"EGL_KHR_surfaceless_context "
// ANNOUNCED BECAUSE IT IS NOW ANSWERED: the four names below are published as real entry
// points and eglGetProcAddress answers them (see Exporting/Definitions.cpp and
// EGLImpl::GetProcAddress). Both earlier states were measured failures: announcing the
// extension while the names resolved to NULL made kwin jump to that NULL (signal 11,
// pc=0x0, from KWin::EglDisplay::determineRenderNode()), and withdrawing the announcement
// left kwin without a render device so its compositing never became active and nothing was
// ever painted.
"EGL_EXT_device_base "
"EGL_EXT_device_enumeration "
"EGL_EXT_device_query "
"EGL_EXT_device_drm "
"EGL_EXT_device_drm_render_node "
// KWin::EglDisplay::determineRenderNode() at libkwin.so.6+0x308b9c. A compositor that is
// never told the extension exists cannot take that path; announcing it again is the right
// move once these four names are really published and resolvable.
"EGL_EXT_platform_device "
"EGL_EXT_platform_base "
"EGL_KHR_platform_base "
"EGL_KHR_platform_gbm "
"EGL_EXT_platform_gbm "
"EGL_MESA_platform_gbm "
"EGL_MESA_platform_surfaceless";
}
@@ -7,6 +7,9 @@
// End of Source File Header
#include <Includes.h>
// EGLImageKHR, for the KHR spellings of the image entry points defined at the bottom of this file:
// Includes.h brings in the core egl.h only.
#include <EGL/eglext.h>
#include "../EGLImpl.h"
MOBILEGL_EGL_API EGLSurface eglCreateWindowSurface(EGLDisplay dpy, EGLConfig config, NativeWindowType window,
@@ -228,6 +231,25 @@ MOBILEGL_EGL_API EGLBoolean eglDestroyImage(EGLDisplay dpy, EGLImage image) {
return MobileGL::MG_Impl::EGLImpl::DestroyImage(dpy, image);
}
// EGL_KHR_image_base's spellings of the same two entry points, exported under their own names
// because that is the only thing that answers for them: measured in the Fedora container that runs
// this library as the glvnd EGL vendor, eglGetProcAddress("eglCreateImageKHR") and
// eglGetProcAddress("eglDestroyImageKHR") both returned NULL, and kwin_wayland's anland backend
// imports the display daemon's dma-bufs through exactly these two.
MOBILEGL_EGL_API EGLImageKHR eglCreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target, EGLClientBuffer buffer,
const EGLint* attrib_list) {
MGLOG_D("eglCreateImageKHR(dpy=%p, ctx=%p, target=%u, buffer=%p, attrib_list=%p)", dpy, ctx, target, buffer,
attrib_list);
// Not an alias of eglCreateImage: this list is EGLint pairs and the core one EGLAttrib pairs.
return MobileGL::MG_Impl::EGLImpl::CreateImageKHR(dpy, ctx, target, buffer, attrib_list);
}
MOBILEGL_EGL_API EGLBoolean eglDestroyImageKHR(EGLDisplay dpy, EGLImageKHR image) {
MGLOG_D("eglDestroyImageKHR(dpy=%p, image=%p)", dpy, image);
// One handle under two spellings: EGLImageKHR and EGLImage are the same pointer.
return MobileGL::MG_Impl::EGLImpl::DestroyImage(dpy, image);
}
MOBILEGL_EGL_API EGLDisplay eglGetPlatformDisplay(EGLenum platform, void* native_display,
const EGLAttrib* attrib_list) {
MGLOG_D("eglGetPlatformDisplay(platform=%u, native_display=%p, attrib_list=%p)", platform, native_display,
@@ -250,6 +272,138 @@ MOBILEGL_EGL_API EGLSurface eglCreatePlatformWindowSurface(EGLDisplay dpy, EGLCo
return MobileGL::MG_Impl::EGLImpl::CreatePlatformWindowSurface(dpy, config, native_window, attrib_list);
}
// THE SWAP-AND-SYNC FAMILY, which kwin asks for in EglBackend::init right after the rendering
// context is up and BEFORE it creates any surface: EGL_KHR_fence_sync, EGL_KHR_swap_buffers_with_damage
// and EGL_EXT_buffer_age are checked there, and a compositing mode whose backend init returns false
// is exactly the "Could not fulfill the requested compositing mode in KWIN_COMPOSE: 1" this run
// reports. Each of these names is a spelling of something this library already implements (the core
// eglSwapBuffers and the state layer's sync objects), so they are published as real exports rather
// than promised and not delivered.
MOBILEGL_EGL_API EGLBoolean eglSwapBuffersWithDamageKHR(EGLDisplay dpy, EGLSurface surface, const EGLint* rects,
EGLint n_rects) {
// The damage rectangles are a hint about what changed; this side swaps the whole surface either
// way, so the honest answer is the core swap.
(void)rects;
(void)n_rects;
return MobileGL::MG_Impl::EGLImpl::SwapBuffers(dpy, surface) ? EGL_TRUE : EGL_FALSE;
}
MOBILEGL_EGL_API EGLBoolean eglSwapBuffersWithDamageEXT(EGLDisplay dpy, EGLSurface surface, const EGLint* rects,
EGLint n_rects) {
return eglSwapBuffersWithDamageKHR(dpy, surface, rects, n_rects);
}
MOBILEGL_EGL_API EGLSyncKHR eglCreateSyncKHR(EGLDisplay dpy, EGLenum type, const EGLint* attrib_list) {
// EGL_KHR_fence_sync's list is EGLint pairs; the state layer takes EGLAttrib pairs.
EGLAttrib attributes[16];
int count = 0;
if (attrib_list != nullptr) {
for (int index = 0; attrib_list[index] != EGL_NONE && count + 1 < 16; index += 2) {
attributes[count++] = static_cast<EGLAttrib>(attrib_list[index]);
attributes[count++] = static_cast<EGLAttrib>(attrib_list[index + 1]);
}
}
attributes[count] = EGL_NONE;
return reinterpret_cast<EGLSyncKHR>(MobileGL::MG_Impl::EGLImpl::CreateSync(dpy, type, attributes));
}
MOBILEGL_EGL_API EGLBoolean eglDestroySyncKHR(EGLDisplay dpy, EGLSyncKHR sync) {
return MobileGL::MG_Impl::EGLImpl::DestroySync(dpy, reinterpret_cast<EGLSync>(sync)) ? EGL_TRUE : EGL_FALSE;
}
MOBILEGL_EGL_API EGLint eglClientWaitSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint flags, EGLTimeKHR timeout) {
return MobileGL::MG_Impl::EGLImpl::ClientWaitSync(dpy, reinterpret_cast<EGLSync>(sync), flags,
static_cast<EGLTime>(timeout));
}
MOBILEGL_EGL_API EGLBoolean eglWaitSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint flags) {
return MobileGL::MG_Impl::EGLImpl::WaitSync(dpy, reinterpret_cast<EGLSync>(sync), flags) ? EGL_TRUE : EGL_FALSE;
}
MOBILEGL_EGL_API EGLBoolean eglGetSyncAttribKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint attribute, EGLint* value) {
EGLAttrib answered = EGL_NONE;
if (!MobileGL::MG_Impl::EGLImpl::GetSyncAttrib(dpy, reinterpret_cast<EGLSync>(sync), attribute, &answered)) {
return EGL_FALSE;
}
if (value != nullptr) {
*value = static_cast<EGLint>(answered);
}
return EGL_TRUE;
}
// THE DEVICE-QUERY FAMILY, PUBLISHED AS REAL ENTRY POINTS. It is not academic: with the extension
// announced but these names unresolvable, libepoxy leaves epoxy_eglQueryDisplayAttribEXT NULL and
// kwin, trusting the announcement, jumps to it (measured: signal 11, fault_addr=(nil), pc=0x0 from
// KWin::EglDisplay::determineRenderNode()); with the announcement withdrawn instead, kwin stops
// asking and its compositing never becomes active (measured: org.kde.kwin.Compositing active =
// false), so nothing is ever painted and the display daemon frames are never imported. Publishing
// them is the only answer that leaves kwin both safe and able to find its render device.
namespace {
// One device, one identity - the same shape the vendor side uses (Exporting/GlvndVendor.cpp): a
// pointer to a private object, so a device from somewhere else cannot compare equal by accident.
struct DeviceToken { int unused; };
DeviceToken g_deviceToken;
} // namespace
MOBILEGL_EGL_API EGLBoolean eglQueryDevicesEXT(EGLint max_devices, EGLDeviceEXT* devices, EGLint* num_devices) {
if (num_devices == nullptr) return EGL_FALSE;
*num_devices = 1;
if (devices != nullptr && max_devices >= 1) devices[0] = reinterpret_cast<EGLDeviceEXT>(&g_deviceToken);
return EGL_TRUE;
}
MOBILEGL_EGL_API EGLBoolean eglQueryDeviceAttribEXT(EGLDeviceEXT device, EGLint attribute, EGLAttrib* value) {
if (device != reinterpret_cast<EGLDeviceEXT>(&g_deviceToken) || value == nullptr) return EGL_FALSE;
switch (attribute) {
case EGL_DRM_DEVICE_FILE_EXT:
case EGL_DRM_RENDER_NODE_FILE_EXT:
*value = reinterpret_cast<EGLAttrib>("/dev/dri/renderD128");
return EGL_TRUE;
default:
return EGL_FALSE;
}
}
MOBILEGL_EGL_API const char* eglQueryDeviceStringEXT(EGLDeviceEXT device, EGLint name) {
if (device != reinterpret_cast<EGLDeviceEXT>(&g_deviceToken)) return nullptr;
// THE NODE IS NAMED, AND THAT CHANGED BECAUSE THE MEASUREMENT CHANGED. An empty answer was the
// right shape while this query was unreachable: nothing asked, so nothing depended on it. Now
// that the name resolves, it is exactly what kwin asks on the way into its EGL backend, and the
// fork says out loud what an empty answer costs it - "Error during init of AnlandEglBackend",
// after which it falls back to the plain AnlandBackend, whose compositing never becomes active
// (supportInformation: "LogicalOutput backend: KWin::AnlandBackend" + "Compositing is not
// active"), so nothing is ever painted and the display daemon frames are never imported. The
// path answered is the one the environment configures the compositor with (ANLAND_DRM_DEVICE),
// so the node named here and the node in use are the same node.
if (name == EGL_DRM_DEVICE_FILE_EXT || name == EGL_DRM_RENDER_NODE_FILE_EXT) {
return "/dev/dri/renderD128";
}
if (name == EGL_EXTENSIONS) return "EGL_EXT_device_drm_render_node";
return nullptr;
}
MOBILEGL_EGL_API EGLBoolean eglQueryDisplayAttribEXT(EGLDisplay dpy, EGLint attribute, EGLAttrib* value) {
(void)dpy;
if (value == nullptr) return EGL_FALSE;
if (attribute == EGL_DEVICE_EXT) {
*value = reinterpret_cast<EGLAttrib>(&g_deviceToken);
return EGL_TRUE;
}
return EGL_FALSE;
}
// The EXT spellings of the platform surface entry points: EGL_EXT_platform_base defines them by those
// names, and a dispatcher-based client resolves them by those names. Identical ABI (EGLAttrib list).
MOBILEGL_EGL_API EGLSurface eglCreatePlatformWindowSurfaceEXT(EGLDisplay dpy, EGLConfig config, void* native_window,
const EGLAttrib* attrib_list) {
return eglCreatePlatformWindowSurface(dpy, config, native_window, attrib_list);
}
MOBILEGL_EGL_API EGLSurface eglCreatePlatformPixmapSurfaceEXT(EGLDisplay dpy, EGLConfig config, void* native_pixmap,
const EGLAttrib* attrib_list) {
return eglCreatePlatformPixmapSurface(dpy, config, native_pixmap, attrib_list);
}
MOBILEGL_EGL_API EGLSurface eglCreatePlatformPixmapSurface(EGLDisplay dpy, EGLConfig config, void* native_pixmap,
const EGLAttrib* attrib_list) {
MGLOG_D("eglCreatePlatformPixmapSurface(dpy=%p, config=%p, native_pixmap=%p, attrib_list=%p)", dpy, config,
@@ -0,0 +1,395 @@
// MobileGL - MobileGL/MG_Impl/EGLImpl/Exporting/GlvndVendor.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// MOBILEGL AS A GLVND EGL VENDOR.
//
// On a system with libglvnd, the libEGL.so.1 an application links is not an implementation at all:
// it is a dispatcher, and the implementations are vendor libraries it loads from
// /usr/share/glvnd/egl_vendor.d (or from whatever __EGL_VENDOR_LIBRARY_FILENAMES names) and routes
// every EGL call to. That is the supported way to be the implementation an application actually
// talks to, and on this machine it is the ONLY way that reaches a caller which does not link EGL
// directly: kwin goes through libepoxy, which dlopen()s libEGL.so.1 and dlsyms its entry points at
// startup, so a library injected with LD_PRELOAD is invisible to it however loudly it is named.
//
// What glvnd asks a vendor for is the other half of that ABI, and this file is it: glvnd calls
// __egl_Main once, the library answers with the functions it wants dispatched to it, and - the part
// that decides whether a compositor can use it at all - with the string naming the PLATFORMS it
// serves. A vendor that does not claim EGL_PLATFORM_GBM_KHR is never asked for a GBM display, and
// the caller that asked for one is told there is no usable device: the client sitting in its maps,
// loaded and never called, is exactly what that looks like from the outside.
//
// The entry points below are this library's own (MG_Impl::EGLImpl, the same functions
// Exporting/Definitions.cpp exports), so a vendor call and a direct call cannot drift apart.
#include <Includes.h>
#include <glvnd/libeglabi.h>
// EGLImageKHR, for the KHR spellings the entry-point table below answers: Includes.h brings in the
// core egl.h only.
#include <EGL/eglext.h>
#include "../EGLImpl.h"
#include "../EGLPlatformExtensions.h"
// The KHR spellings of the image entry points, declared here because the bundled EGL headers
// declare them only under EGL_EGLEXT_PROTOTYPES and because glvnd asks for entry points BY NAME -
// the table below is where it finds this one.
extern "C" EGLImageKHR eglCreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target, EGLClientBuffer buffer,
const EGLint* attrib_list);
extern "C" EGLBoolean eglDestroyImageKHR(EGLDisplay dpy, EGLImageKHR image);
// THE SWAP-AND-SYNC EXPORTS, declared here for the same reason the image pair is: they are defined in
// Exporting/Definitions.cpp and this file only names them in the vendor table below.
extern "C" EGLBoolean eglSwapBuffersWithDamageKHR(EGLDisplay dpy, EGLSurface surface, const EGLint* rects,
EGLint n_rects);
extern "C" EGLBoolean eglSwapBuffersWithDamageEXT(EGLDisplay dpy, EGLSurface surface, const EGLint* rects,
EGLint n_rects);
extern "C" EGLSyncKHR eglCreateSyncKHR(EGLDisplay dpy, EGLenum type, const EGLint* attrib_list);
extern "C" EGLBoolean eglDestroySyncKHR(EGLDisplay dpy, EGLSyncKHR sync);
extern "C" EGLint eglClientWaitSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint flags, EGLTimeKHR timeout);
extern "C" EGLBoolean eglWaitSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint flags);
extern "C" EGLBoolean eglGetSyncAttribKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint attribute, EGLint* value);
namespace {
using namespace MobileGL;
const __EGLapiExports* g_eglExports = nullptr;
// The platforms this vendor serves. GBM is the one that matters here: the compositor asks for
// a display on the GBM device it opened, and a vendor that does not name the platform is not
// asked at all. This side treats a GBM device as it treats every other native display - the
// frames come from the display host, not from the device - so claiming the platform is honest
// rather than optimistic.
constexpr const char* kPlatformExtensions = MG_Impl::EGLImpl::kPlatformExtensionString;
// glvnd asks for the client-only extension list separately from the platform one; the split is
// its own invention, and this side answers both from the same query.
const char* EGLAPIENTRY VendorQueryString(EGLDisplay dpy, EGLint name) {
return MG_Impl::EGLImpl::QueryString(dpy, name);
}
EGLDisplay EGLAPIENTRY VendorGetPlatformDisplay(EGLenum platform, void* native_display,
const EGLAttrib* attrib_list) {
// EGL_NONE means the application called eglGetDisplay(EGL_DEFAULT_DISPLAY): the vendor is
// being asked to name its own default display.
if (platform == EGL_NONE) {
return MG_Impl::EGLImpl::GetDisplay(EGL_DEFAULT_DISPLAY);
}
return MG_Impl::EGLImpl::GetPlatformDisplay(platform, native_display, attrib_list);
}
EGLBoolean EGLAPIENTRY VendorGetSupportsAPI(EGLenum api) {
// What this library answers for EGL_CLIENT_APIS, said as a question.
return (api == EGL_OPENGL_ES_API || api == EGL_OPENGL_API) ? EGL_TRUE : EGL_FALSE;
}
const char* EGLAPIENTRY VendorGetVendorString(int name) {
if (name == __EGL_VENDOR_STRING_PLATFORM_EXTENSIONS) {
return kPlatformExtensions;
}
return nullptr;
}
// GLVND ASKS FOR EVERY EGL ENTRY POINT BY NAME, AND DROPS THE VENDOR IF ONE IS MISSING: the
// first eight functions the table below names are looked up before this library is ever asked
// for a display, and a single null answer means glvnd quietly stops considering it - the
// library is loaded, __egl_Main has run, and nothing is ever dispatched to it.
struct EntryPoint {
const char* name;
void* address;
};
// ---- the device family -------------------------------------------------------------------
//
// A compositor does not ask for a display and hope. It asks which devices exist, reads the DRM
// node off the one it wants, and only then asks for a display ON that device - so a vendor that
// answers "no devices" is one it cannot use, however well the vendor draws. The message it
// gives is "no usable DRM render device", which names where it stopped rather than why.
//
// This implementation therefore names one device: the render node of the machine it runs on.
// That is an honest answer even though the frames do not come from that node - they come from
// the display host on the other end of the socket, which is why every surface here is a host
// frame - because the question is which device this display is associated with, and the answer
// is what lets the compositor match its own node to the display it is about to draw into.
const char* DrmRenderNode() {
// MOBILEGL_DEVICE_DRM_NODE names another node: a machine with more than one GPU, or a test
// that wants the mismatch refused rather than papered over.
const char* configured = std::getenv("MOBILEGL_DEVICE_DRM_NODE");
return (configured != nullptr && configured[0] != '\0') ? configured : "/dev/dri/renderD128";
}
// One device, one identity. A pointer to a private object rather than a number, so that a
// device from somewhere else cannot compare equal to this one by accident.
struct DeviceToken { int unused; };
DeviceToken g_deviceToken;
EGLBoolean EGLAPIENTRY VendorQueryDevicesEXT(EGLint max_devices, EGLDeviceEXT* devices, EGLint* num_devices) {
if (num_devices == nullptr) {
return EGL_FALSE;
}
*num_devices = 1;
if (devices != nullptr && max_devices >= 1) {
devices[0] = reinterpret_cast<EGLDeviceEXT>(&g_deviceToken);
}
return EGL_TRUE;
}
EGLBoolean EGLAPIENTRY VendorQueryDeviceAttribEXT(EGLDeviceEXT device, EGLint attribute, EGLAttrib* value) {
MGLOG_I("vendor eglQueryDeviceAttribEXT(device=%p, attribute=0x%04x)", static_cast<void*>(device),
static_cast<unsigned>(attribute));
if (device != reinterpret_cast<EGLDeviceEXT>(&g_deviceToken) || value == nullptr) {
return EGL_FALSE;
}
switch (attribute) {
case EGL_DRM_DEVICE_FILE_EXT:
case EGL_DRM_RENDER_NODE_FILE_EXT:
// THE ATTRIBUTE FORM OF THE SAME QUERY, and the last place the empty answer lived. kwin
// asks for the node as an ATTRIBUTE as well as a string; with this one empty its log said
// "couldn't find dev node for drm device" even after the string form was answered, and
// the compositing mode it had just successfully started through this library was then
// declared unfulfillable. The path is the environment's own (ANLAND_DRM_DEVICE).
*value = reinterpret_cast<EGLAttrib>("/dev/dri/renderD128");
return EGL_TRUE;
default:
return EGL_FALSE;
}
}
const char* EGLAPIENTRY VendorQueryDeviceStringEXT(EGLDeviceEXT device, EGLint name) {
// WHICH DEVICE QUERY THE COMPOSITOR MAKES, and by which name: the string form and the attribute
// form are different entry points with different answers (glvnd answers the string form itself
// and only forwards the attribute one), so the caller decides which of the two it gets.
MGLOG_I("vendor eglQueryDeviceStringEXT(device=%p, name=0x%04x)", static_cast<void*>(device),
static_cast<unsigned>(name));
if (device != reinterpret_cast<EGLDeviceEXT>(&g_deviceToken)) {
return nullptr;
}
// THE NODE IS NAMED HERE TOO, AND THAT IS WHAT THE VENDOR-SIDE ANSWER WAS MISSING. An empty
// answer was measured to be harmless only while this query was unreachable; through glvnd it
// IS the query a compositor makes before it opens anything, and an empty string is what its
// next line reports - "Failed to open drm node : No such file or directory" (the name is the
// blank between "node" and ":"), followed by "couldn't find dev node for drm device" and a
// dead OpenGL compositing mode. The path is the one the environment configures the
// compositor with (ANLAND_DRM_DEVICE), so the node named here and the node in use are one.
if (name == EGL_DRM_DEVICE_FILE_EXT || name == EGL_DRM_RENDER_NODE_FILE_EXT) {
return "/dev/dri/renderD128";
}
if (name == EGL_EXTENSIONS) {
return "EGL_EXT_device_drm_render_node";
}
return nullptr;
}
EGLBoolean EGLAPIENTRY VendorQueryDisplayAttribEXT(EGLDisplay dpy, EGLint attribute, EGLAttrib* value) {
(void)dpy;
if (value == nullptr) {
return EGL_FALSE;
}
if (attribute == EGL_DEVICE_EXT) {
*value = reinterpret_cast<EGLAttrib>(&g_deviceToken);
return EGL_TRUE;
}
return EGL_FALSE;
}
// Three of the names glvnd insists on are debug extensions this implementation does not
// provide, and glvnd insists on them all the same - a vendor that answers null for any of them
// is dropped. They are answered honestly rather than pretended: KHR_debug is not in the
// extension strings above, so a caller that reads them will not call these at all.
EGLint EGLAPIENTRY VendorDebugMessageControlKHR(EGLDEBUGPROCKHR callback, const EGLAttrib* attrib_list) {
(void)callback;
(void)attrib_list;
return EGL_FALSE;
}
EGLBoolean EGLAPIENTRY VendorQueryDebugKHR(EGLint attribute, EGLAttrib* value) {
(void)attribute;
(void)value;
return EGL_FALSE;
}
EGLint EGLAPIENTRY VendorLabelObjectKHR(EGLDisplay dpy, EGLenum objectType, EGLObjectKHR object, EGLLabelKHR label) {
(void)dpy;
(void)objectType;
(void)object;
(void)label;
return EGL_BAD_PARAMETER;
}
const EntryPoint kEglEntryPoints[] = {
// The core of EGL 1.5, every one of which glvnd looks up by name before it considers this
// library usable.
{"eglInitialize", reinterpret_cast<void*>(eglInitialize)},
{"eglChooseConfig", reinterpret_cast<void*>(eglChooseConfig)},
{"eglCopyBuffers", reinterpret_cast<void*>(eglCopyBuffers)},
{"eglCreateContext", reinterpret_cast<void*>(eglCreateContext)},
{"eglCreatePbufferSurface", reinterpret_cast<void*>(eglCreatePbufferSurface)},
{"eglCreatePixmapSurface", reinterpret_cast<void*>(eglCreatePixmapSurface)},
{"eglCreateWindowSurface", reinterpret_cast<void*>(eglCreateWindowSurface)},
{"eglDestroyContext", reinterpret_cast<void*>(eglDestroyContext)},
{"eglDestroySurface", reinterpret_cast<void*>(eglDestroySurface)},
{"eglGetConfigAttrib", reinterpret_cast<void*>(eglGetConfigAttrib)},
{"eglGetConfigs", reinterpret_cast<void*>(eglGetConfigs)},
{"eglMakeCurrent", reinterpret_cast<void*>(eglMakeCurrent)},
{"eglQueryContext", reinterpret_cast<void*>(eglQueryContext)},
{"eglQuerySurface", reinterpret_cast<void*>(eglQuerySurface)},
{"eglSwapBuffers", reinterpret_cast<void*>(eglSwapBuffers)},
{"eglTerminate", reinterpret_cast<void*>(eglTerminate)},
{"eglWaitGL", reinterpret_cast<void*>(eglWaitGL)},
{"eglWaitNative", reinterpret_cast<void*>(eglWaitNative)},
{"eglBindTexImage", reinterpret_cast<void*>(eglBindTexImage)},
{"eglReleaseTexImage", reinterpret_cast<void*>(eglReleaseTexImage)},
{"eglSurfaceAttrib", reinterpret_cast<void*>(eglSurfaceAttrib)},
{"eglSwapInterval", reinterpret_cast<void*>(eglSwapInterval)},
{"eglBindAPI", reinterpret_cast<void*>(eglBindAPI)},
{"eglCreatePbufferFromClientBuffer", reinterpret_cast<void*>(eglCreatePbufferFromClientBuffer)},
{"eglReleaseThread", reinterpret_cast<void*>(eglReleaseThread)},
{"eglWaitClient", reinterpret_cast<void*>(eglWaitClient)},
{"eglGetError", reinterpret_cast<void*>(eglGetError)},
{"eglCreateSync", reinterpret_cast<void*>(eglCreateSync)},
{"eglDestroySync", reinterpret_cast<void*>(eglDestroySync)},
{"eglClientWaitSync", reinterpret_cast<void*>(eglClientWaitSync)},
{"eglGetSyncAttrib", reinterpret_cast<void*>(eglGetSyncAttrib)},
{"eglWaitSync", reinterpret_cast<void*>(eglWaitSync)},
{"eglCreateImage", reinterpret_cast<void*>(eglCreateImage)},
{"eglDestroyImage", reinterpret_cast<void*>(eglDestroyImage)},
// THE KHR SPELLINGS OF THOSE TWO, and they are entries rather than aliases because glvnd
// resolves a name it does not know through this table alone. Measured in the Fedora
// container that runs this library as the EGL vendor: with only the core pair here,
// eglGetProcAddress("eglCreateImageKHR") and eglGetProcAddress("eglDestroyImageKHR") both
// answered NULL, and kwin_wayland's anland backend reaches the display daemon's dma-bufs
// through them (AnlandEglLayer::importBuffers -> EglBackend::importDmaBufAsTexture ->
// EglDisplay::importBufferAsImage) - so the compositor had no entry point to import a frame
// with. eglCreateImageKHR is not the same ABI as eglCreateImage: EGL_KHR_image_base's
// attribute list is EGLint pairs, which is why Exporting/Definitions.cpp exports it as its
// own function rather than letting the core spelling stand in for it.
{"eglCreateImageKHR", reinterpret_cast<void*>(eglCreateImageKHR)},
{"eglDestroyImageKHR", reinterpret_cast<void*>(eglDestroyImageKHR)},
{"eglCreatePlatformWindowSurface", reinterpret_cast<void*>(eglCreatePlatformWindowSurface)},
{"eglCreatePlatformPixmapSurface", reinterpret_cast<void*>(eglCreatePlatformPixmapSurface)},
// EGL_EXT_platform_base's OWN SPELLINGS, and the reason this file exists at all: the vendor
// string below advertises that extension, whose entry points ARE the EXT-named ones, and a
// dispatcher-based client resolves them by exactly those names (libepoxy in kwin does). With
// only the core spellings in this table the lookup answered NULL and the caller - which had
// already checked the extension string and so had every reason to trust it - jumped to it:
// measured as signal 11 with fault_addr=(nil) and pc=0x0 immediately after "MobileGL
// initialized", in both the systemd and the direct run, and it is what kept the compositor
// from ever reaching its output. The ABI is identical (an EGLAttrib list, not EGLint pairs),
// so the core implementation is what serves them.
{"eglCreatePlatformWindowSurfaceEXT", reinterpret_cast<void*>(eglCreatePlatformWindowSurface)},
{"eglCreatePlatformPixmapSurfaceEXT", reinterpret_cast<void*>(eglCreatePlatformPixmapSurface)},
// The platform and current-state queries: not in glvnd's required list, but the names an
// application that talks to a vendor through a dispatcher will ask for by name.
{"eglGetDisplay", reinterpret_cast<void*>(eglGetDisplay)},
{"eglGetCurrentDisplay", reinterpret_cast<void*>(eglGetCurrentDisplay)},
{"eglGetCurrentSurface", reinterpret_cast<void*>(eglGetCurrentSurface)},
{"eglGetCurrentContext", reinterpret_cast<void*>(eglGetCurrentContext)},
{"eglQueryAPI", reinterpret_cast<void*>(eglQueryAPI)},
{"eglGetProcAddress", reinterpret_cast<void*>(eglGetProcAddress)},
// The two queries whose answer depends on which side of glvnd is asking.
{"eglQueryString", reinterpret_cast<void*>(VendorQueryString)},
{"eglGetPlatformDisplay", reinterpret_cast<void*>(VendorGetPlatformDisplay)},
{"eglGetPlatformDisplayEXT", reinterpret_cast<void*>(VendorGetPlatformDisplay)},
// What glvnd requires and this implementation does not otherwise have.
{"eglQueryDevicesEXT", reinterpret_cast<void*>(VendorQueryDevicesEXT)},
{"eglQueryDeviceAttribEXT", reinterpret_cast<void*>(VendorQueryDeviceAttribEXT)},
{"eglQueryDeviceStringEXT", reinterpret_cast<void*>(VendorQueryDeviceStringEXT)},
{"eglQueryDisplayAttribEXT", reinterpret_cast<void*>(VendorQueryDisplayAttribEXT)},
// The swap-and-sync family, published through the same table the names are answered from.
// Declared above this table because they are exports of Definitions.cpp, like the image pair.
{"eglSwapBuffersWithDamageKHR", reinterpret_cast<void*>(eglSwapBuffersWithDamageKHR)},
{"eglSwapBuffersWithDamageEXT", reinterpret_cast<void*>(eglSwapBuffersWithDamageEXT)},
{"eglCreateSyncKHR", reinterpret_cast<void*>(eglCreateSyncKHR)},
{"eglDestroySyncKHR", reinterpret_cast<void*>(eglDestroySyncKHR)},
{"eglClientWaitSyncKHR", reinterpret_cast<void*>(eglClientWaitSyncKHR)},
{"eglWaitSyncKHR", reinterpret_cast<void*>(eglWaitSyncKHR)},
{"eglGetSyncAttribKHR", reinterpret_cast<void*>(eglGetSyncAttribKHR)},
{"eglDebugMessageControlKHR", reinterpret_cast<void*>(VendorDebugMessageControlKHR)},
{"eglQueryDebugKHR", reinterpret_cast<void*>(VendorQueryDebugKHR)},
{"eglLabelObjectKHR", reinterpret_cast<void*>(VendorLabelObjectKHR)},
};
// THE SAME ANSWER TO BOTH OF GLVND'S QUESTIONS, and that is not a shortcut - glvnd asks them
// from two places and a library that answers one of them has a hole the size of the other:
//
// getProcAddress is asked for every entry point glvnd builds its own static EGL table
// from, so a null here drops the vendor outright.
// getDispatchAddress is what glvnd falls back to for eglGetProcAddress on any name it does
// not know itself - which is every extension entry point that is not in
// its list, the device queries among them. A vendor that answers these
// from a GL-only table hands the application a null for a function the
// vendor implements, and the caller reports the extension as missing.
void* LookupEntryPoint(const char* procName) {
if (procName == nullptr) {
return nullptr;
}
for (const EntryPoint& entry : kEglEntryPoints) {
if (std::strcmp(entry.name, procName) == 0) {
return entry.address;
}
}
// Anything that is not EGL is a GL or GLES entry point, and the tree's own resolver is the
// table for those.
return reinterpret_cast<void*>(MG_Impl::EGLImpl::GetProcAddress(procName));
}
void* EGLAPIENTRY VendorGetProcAddress(const char* procName) { return LookupEntryPoint(procName); }
// glvnd hands a vendor the index each EGL entry point occupies in its dispatch table, so that a
// vendor whose entry points are thin wrappers around glvnd can call back into the right slot.
// This library's entry points ARE the implementation - they never re-enter glvnd - so there is
// nothing to record; the callback exists because glvnd refuses a vendor that does not provide
// one, and refusing it means the library is loaded and never dispatched to.
void EGLAPIENTRY VendorSetDispatchIndex(const char* procName, int index) {
(void)procName;
(void)index;
}
void* EGLAPIENTRY VendorGetDispatchAddress(const char* procName) {
// glvnd asks THIS vendor for GL entry points as well: after eglMakeCurrent, the dispatch
// table of the now-current context is filled from here, one name at a time. LookupEntryPoint
// knows EGL names only, so every GL name (glGetString, glCreateProgram, ...) answered NULL,
// glvnd wrote NULL into its dispatch slots, and the first glGetString of the compositor's
// context came back NULL - kwin's "Could not initialize rendering context", while the very
// same context answered fine through a direct eglGetProcAddress (measured: probe6 vs probe12).
// GL names go through the full GL/EGL lookup, which is the same table eglGetProcAddress uses.
if (void* address = LookupEntryPoint(procName)) {
return address;
}
return reinterpret_cast<void*>(MG_Impl::EGLImpl::GetProcAddress(procName));
}
} // namespace
extern "C" MOBILEGL_EGL_API EGLBoolean __egl_Main(uint32_t version, const __EGLapiExports* exports,
__EGLvendorInfo* vendor, __EGLapiImports* imports) {
(void)vendor;
if (EGL_VENDOR_ABI_GET_MAJOR_VERSION(version) != EGL_VENDOR_ABI_MAJOR_VERSION) {
// A vendor that does not know the ABI it is being handed must say so, not guess: the
// structures on either side are the ABI.
MGLOG_E("__egl_Main: unsupported glvnd vendor ABI version 0x%08x (this library implements %u.%u)", version,
static_cast<unsigned>(EGL_VENDOR_ABI_MAJOR_VERSION),
static_cast<unsigned>(EGL_VENDOR_ABI_MINOR_VERSION));
return EGL_FALSE;
}
if (exports == nullptr || imports == nullptr) {
MGLOG_E("__egl_Main: glvnd handed no exports/imports table");
return EGL_FALSE;
}
g_eglExports = exports;
imports->getPlatformDisplay = VendorGetPlatformDisplay;
imports->getSupportsAPI = VendorGetSupportsAPI;
imports->getVendorString = VendorGetVendorString;
imports->getProcAddress = VendorGetProcAddress;
imports->getDispatchAddress = VendorGetDispatchAddress;
imports->setDispatchIndex = VendorSetDispatchIndex;
MGLOG_I("glvnd vendor: MobileGL is serving EGL through libglvnd (ABI %u.%u, platforms: %s)",
static_cast<unsigned>(EGL_VENDOR_ABI_MAJOR_VERSION), static_cast<unsigned>(EGL_VENDOR_ABI_MINOR_VERSION),
kPlatformExtensions);
return EGL_TRUE;
}
+398
View File
@@ -0,0 +1,398 @@
// MobileGL - MobileGL/MG_Impl/EGLImpl/WaylandWindow.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#include "WaylandWindow.h"
#if MOBILEGL_WAYLAND_WINDOWS
#include <MG_Impl/GLImpl/Buffer/GL_Buffer.h>
#include <MG_Impl/GLImpl/Framebuffer/GL_Framebuffer.h>
#include <MG_Impl/GLImpl/Getter/GL_Getter.h>
#include <MG_Impl/GLImpl/RenderState/GL_RenderState.h>
#include <MG_Util/Debug/Log.h>
#include <cerrno>
#include <cstdarg>
#include <cstdint>
#include <cstring>
#include <dlfcn.h>
#include <fcntl.h>
#include <mutex>
#include <sys/mman.h>
#include <unistd.h>
namespace MobileGL::MG_Impl::EGLImpl::Wayland {
namespace {
// libwayland-client's types are opaque here: everything goes through wl_proxy_* and the
// exported interface descriptions, and nothing reads a field of either.
struct wl_proxy;
struct wl_interface;
struct wl_event_queue;
// wayland-egl-backend.h, WL_EGL_WINDOW_VERSION 3: the struct libwayland-egl allocates for
// wl_egl_window_create and the driver reads. The ABI is versioned by its first word.
struct EglWindow {
const intptr_t version;
int width;
int height;
int dx;
int dy;
int attached_width;
int attached_height;
void* driver_private;
void (*resize_callback)(EglWindow*, void*);
void (*destroy_window_callback)(void*);
wl_proxy* surface;
};
constexpr intptr_t kEglWindowVersion = 3;
// Request opcodes (wayland.xml, the order the requests are declared in).
constexpr uint32_t kDisplayGetRegistry = 1;
constexpr uint32_t kRegistryBind = 0;
constexpr uint32_t kShmCreatePool = 0;
constexpr uint32_t kShmPoolCreateBuffer = 0;
constexpr uint32_t kShmPoolDestroy = 1;
constexpr uint32_t kBufferDestroy = 0;
constexpr uint32_t kSurfaceAttach = 1;
constexpr uint32_t kSurfaceDamage = 2;
constexpr uint32_t kSurfaceCommit = 6;
constexpr uint32_t kSurfaceDamageBuffer = 9; // since wl_surface version 4
constexpr uint32_t kMarshalFlagDestroy = 1u << 0;
// wl_shm formats. The two every compositor must support; byte order in memory is
// B, G, R, A for both (little-endian 0xAARRGGBB).
constexpr uint32_t kShmFormatArgb8888 = 0;
constexpr uint32_t kShmFormatXrgb8888 = 1;
struct Api {
Bool loaded = false;
const wl_interface* displayInterface = nullptr;
const wl_interface* registryInterface = nullptr;
const wl_interface* shmInterface = nullptr;
const wl_interface* shmPoolInterface = nullptr;
const wl_interface* bufferInterface = nullptr;
wl_proxy* (*marshalFlags)(wl_proxy*, uint32_t, const wl_interface*, uint32_t, uint32_t, ...) = nullptr;
int (*addListener)(wl_proxy*, void (**)(void), void*) = nullptr;
void (*destroy)(wl_proxy*) = nullptr;
uint32_t (*getVersion)(wl_proxy*) = nullptr;
void* (*createWrapper)(void*) = nullptr;
void (*wrapperDestroy)(void*) = nullptr;
void (*setQueue)(wl_proxy*, wl_event_queue*) = nullptr;
wl_event_queue* (*createQueue)(void*) = nullptr;
void (*queueDestroy)(wl_event_queue*) = nullptr;
int (*roundtripQueue)(void*, wl_event_queue*) = nullptr;
int (*dispatchQueue)(void*, wl_event_queue*) = nullptr;
int (*dispatchQueuePending)(void*, wl_event_queue*) = nullptr;
int (*flush)(void*) = nullptr;
};
// Loaded once. RTLD_NOLOAD first: a Wayland client has the library already, and the copy
// it is using is the one whose wl_display this must talk to.
const Api& WaylandApi() {
static Api api;
static std::once_flag once;
std::call_once(once, [] {
void* lib = dlopen("libwayland-client.so.0", RTLD_NOW | RTLD_NOLOAD);
if (lib == nullptr) lib = dlopen("libwayland-client.so.0", RTLD_NOW);
if (lib == nullptr) return;
const auto sym = [lib](const char* name) { return dlsym(lib, name); };
api.displayInterface = static_cast<const wl_interface*>(sym("wl_display_interface"));
api.registryInterface = static_cast<const wl_interface*>(sym("wl_registry_interface"));
api.shmInterface = static_cast<const wl_interface*>(sym("wl_shm_interface"));
api.shmPoolInterface = static_cast<const wl_interface*>(sym("wl_shm_pool_interface"));
api.bufferInterface = static_cast<const wl_interface*>(sym("wl_buffer_interface"));
api.marshalFlags = reinterpret_cast<decltype(api.marshalFlags)>(sym("wl_proxy_marshal_flags"));
api.addListener = reinterpret_cast<decltype(api.addListener)>(sym("wl_proxy_add_listener"));
api.destroy = reinterpret_cast<decltype(api.destroy)>(sym("wl_proxy_destroy"));
api.getVersion = reinterpret_cast<decltype(api.getVersion)>(sym("wl_proxy_get_version"));
api.createWrapper = reinterpret_cast<decltype(api.createWrapper)>(sym("wl_proxy_create_wrapper"));
api.wrapperDestroy = reinterpret_cast<decltype(api.wrapperDestroy)>(sym("wl_proxy_wrapper_destroy"));
api.setQueue = reinterpret_cast<decltype(api.setQueue)>(sym("wl_proxy_set_queue"));
api.createQueue = reinterpret_cast<decltype(api.createQueue)>(sym("wl_display_create_queue"));
api.queueDestroy = reinterpret_cast<decltype(api.queueDestroy)>(sym("wl_event_queue_destroy"));
api.roundtripQueue = reinterpret_cast<decltype(api.roundtripQueue)>(sym("wl_display_roundtrip_queue"));
api.dispatchQueue = reinterpret_cast<decltype(api.dispatchQueue)>(sym("wl_display_dispatch_queue"));
api.dispatchQueuePending =
reinterpret_cast<decltype(api.dispatchQueuePending)>(sym("wl_display_dispatch_queue_pending"));
api.flush = reinterpret_cast<decltype(api.flush)>(sym("wl_display_flush"));
api.loaded = api.displayInterface && api.registryInterface && api.shmInterface &&
api.shmPoolInterface && api.bufferInterface && api.marshalFlags && api.addListener && api.destroy &&
api.getVersion && api.createWrapper && api.wrapperDestroy && api.setQueue &&
api.createQueue && api.queueDestroy && api.roundtripQueue && api.dispatchQueue &&
api.dispatchQueuePending && api.flush;
if (!api.loaded) MGLOG_E("Wayland: libwayland-client.so.0 lacks an entry point this library uses");
});
return api;
}
struct ShmBuffer {
wl_proxy* buffer = nullptr;
void* map = nullptr;
SizeT size = 0;
EGLint width = 0;
EGLint height = 0;
EGLint stride = 0;
Bool busy = false;
};
// The compositor is done with a buffer: it may be drawn into again.
void OnBufferRelease(void* data, wl_proxy*) { static_cast<ShmBuffer*>(data)->busy = false; }
void (*const kBufferListener[])(void) = {reinterpret_cast<void (*)(void)>(&OnBufferRelease)};
} // namespace
struct WindowSurface::Impl {
void* display = nullptr;
EglWindow* window = nullptr;
wl_event_queue* queue = nullptr;
void* displayWrapper = nullptr;
wl_proxy* registry = nullptr;
wl_proxy* shm = nullptr;
uint32_t format = kShmFormatXrgb8888;
EGLint width = 0;
EGLint height = 0;
// Three: one on the glass, one the compositor may still be reading, one to draw into.
ShmBuffer buffers[3];
Vector<Uint8> scratch;
~Impl() {
const Api& api = WaylandApi();
for (auto& buffer : buffers) Release(buffer);
// Neither wl_shm (version 1) nor wl_registry has a destroy request: the proxies go
// on this side only.
if (shm) api.destroy(shm);
if (registry) api.destroy(registry);
if (displayWrapper) api.wrapperDestroy(displayWrapper);
if (queue) api.queueDestroy(queue);
if (window && window->driver_private == this) {
window->driver_private = nullptr;
window->destroy_window_callback = nullptr;
}
}
void Release(ShmBuffer& buffer) {
const Api& api = WaylandApi();
if (buffer.buffer) api.marshalFlags(buffer.buffer, kBufferDestroy, nullptr, api.getVersion(buffer.buffer),
kMarshalFlagDestroy);
if (buffer.map) munmap(buffer.map, buffer.size);
buffer = ShmBuffer{};
}
Bool Allocate(ShmBuffer& buffer) {
const Api& api = WaylandApi();
Release(buffer);
const EGLint stride = width * 4;
const SizeT size = static_cast<SizeT>(stride) * static_cast<SizeT>(height);
const int fd = memfd_create("mobilegl-wayland", MFD_CLOEXEC);
if (fd < 0) {
MGLOG_E("Wayland: memfd_create for a %dx%d buffer failed: %s", width, height, std::strerror(errno));
return false;
}
if (ftruncate(fd, static_cast<off_t>(size)) != 0) {
MGLOG_E("Wayland: sizing a %dx%d buffer failed: %s", width, height, std::strerror(errno));
close(fd);
return false;
}
void* map = mmap(nullptr, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
if (map == MAP_FAILED) {
MGLOG_E("Wayland: mapping a %dx%d buffer failed: %s", width, height, std::strerror(errno));
close(fd);
return false;
}
// One pool per buffer, destroyed at once: the buffer keeps the memory alive.
wl_proxy* pool = api.marshalFlags(shm, kShmCreatePool, api.shmPoolInterface, api.getVersion(shm), 0, nullptr,
fd, static_cast<int32_t>(size));
close(fd);
if (pool == nullptr) {
munmap(map, size);
return false;
}
wl_proxy* wlBuffer = api.marshalFlags(pool, kShmPoolCreateBuffer, api.bufferInterface, api.getVersion(pool),
0, nullptr, 0, width, height, stride, format);
api.marshalFlags(pool, kShmPoolDestroy, nullptr, api.getVersion(pool), kMarshalFlagDestroy);
if (wlBuffer == nullptr) {
munmap(map, size);
return false;
}
buffer.buffer = wlBuffer;
buffer.map = map;
buffer.size = size;
buffer.width = width;
buffer.height = height;
buffer.stride = stride;
buffer.busy = false;
api.addListener(wlBuffer, const_cast<void (**)(void)>(kBufferListener), &buffer);
return true;
}
// A buffer the compositor is not reading: the events already received first, then (all
// three still held) a blocking wait for the next release.
ShmBuffer* NextFree() {
const Api& api = WaylandApi();
for (;;) {
if (api.dispatchQueuePending(display, queue) < 0) return nullptr;
for (auto& buffer : buffers) {
if (buffer.buffer == nullptr || buffer.busy) continue;
if (buffer.width != width || buffer.height != height) {
if (!Allocate(buffer)) return nullptr;
}
return &buffer;
}
for (auto& buffer : buffers) {
if (buffer.buffer == nullptr) return Allocate(buffer) ? &buffer : nullptr;
}
if (api.dispatchQueue(display, queue) < 0) return nullptr;
}
}
};
namespace {
void OnRegistryGlobal(void* data, wl_proxy* registry, uint32_t name, const char* interface, uint32_t) {
auto* impl = static_cast<WindowSurface::Impl*>(data);
if (impl->shm != nullptr || std::strcmp(interface, "wl_shm") != 0) return;
const Api& api = WaylandApi();
impl->shm = api.marshalFlags(registry, kRegistryBind, api.shmInterface, 1, 0, name, "wl_shm", 1u, nullptr);
}
void OnRegistryGlobalRemove(void*, wl_proxy*, uint32_t) {}
void (*const kRegistryListener[])(void) = {reinterpret_cast<void (*)(void)>(&OnRegistryGlobal),
reinterpret_cast<void (*)(void)>(&OnRegistryGlobalRemove)};
// The application destroyed its wl_egl_window first: nothing may be read from it after this.
void OnWindowDestroyed(void* data) { static_cast<WindowSurface::Impl*>(data)->window = nullptr; }
} // namespace
Bool IsWaylandDisplay(Uint64 nativeDisplay, EGLenum platform) {
if (platform == kPlatformWayland) return true;
if (nativeDisplay == 0 || (platform != EGL_NONE && platform != 0)) return false;
const Api& api = WaylandApi();
if (!api.loaded) return false;
return *reinterpret_cast<const void* const*>(static_cast<uintptr_t>(nativeDisplay)) == api.displayInterface;
}
Bool IsWaylandWindow(const void* window) {
if (window == nullptr) return false;
const auto* eglWindow = static_cast<const EglWindow*>(window);
return eglWindow->version >= kEglWindowVersion && eglWindow->surface != nullptr;
}
Bool WindowSize(const void* window, EGLint* width, EGLint* height) {
if (!IsWaylandWindow(window)) return false;
const auto* eglWindow = static_cast<const EglWindow*>(window);
*width = eglWindow->width;
*height = eglWindow->height;
return eglWindow->width > 0 && eglWindow->height > 0;
}
WindowSurface::WindowSurface(UniquePtr<Impl> impl) : m_impl(std::move(impl)) {}
WindowSurface::~WindowSurface() = default;
UniquePtr<WindowSurface> WindowSurface::Create(void* wlDisplay, void* wlEglWindow, EGLint width, EGLint height,
Bool hasAlpha) {
const Api& api = WaylandApi();
if (!api.loaded) {
MGLOG_E("Wayland: a window surface needs libwayland-client.so.0, and it could not be used");
return nullptr;
}
if (wlDisplay == nullptr || !IsWaylandWindow(wlEglWindow) || width <= 0 || height <= 0) return nullptr;
auto impl = MakeUnique<Impl>();
impl->display = wlDisplay;
impl->window = static_cast<EglWindow*>(wlEglWindow);
impl->width = width;
impl->height = height;
impl->format = hasAlpha ? kShmFormatArgb8888 : kShmFormatXrgb8888;
// A queue of this surface's own, so the application's dispatching never sees these events
// and this never dispatches the application's.
impl->queue = api.createQueue(wlDisplay);
impl->displayWrapper = api.createWrapper(wlDisplay);
if (impl->queue == nullptr || impl->displayWrapper == nullptr) return nullptr;
api.setQueue(static_cast<wl_proxy*>(impl->displayWrapper), impl->queue);
auto* wrapper = static_cast<wl_proxy*>(impl->displayWrapper);
impl->registry = api.marshalFlags(wrapper, kDisplayGetRegistry, api.registryInterface, api.getVersion(wrapper),
0, nullptr);
if (impl->registry == nullptr) return nullptr;
api.addListener(impl->registry, const_cast<void (**)(void)>(kRegistryListener), impl.get());
if (api.roundtripQueue(wlDisplay, impl->queue) < 0 || impl->shm == nullptr) {
MGLOG_E("Wayland: the compositor offers no wl_shm; a window surface cannot be presented");
return nullptr;
}
impl->window->driver_private = impl.get();
impl->window->destroy_window_callback = &OnWindowDestroyed;
MGLOG_I("Wayland: window surface %dx%d presented through wl_shm (%s)", width, height,
hasAlpha ? "ARGB8888" : "XRGB8888");
return UniquePtr<WindowSurface>(new WindowSurface(std::move(impl)));
}
Bool WindowSurface::Present() {
Impl& impl = *m_impl;
if (impl.window == nullptr) return false;
const Api& api = WaylandApi();
ShmBuffer* buffer = impl.NextFree();
if (buffer == nullptr) {
MGLOG_E_ONCE("Wayland: no wl_shm buffer could be taken for the frame; it is not shown");
return false;
}
// The frame, out of the default framebuffer, through this library's own entry points: the
// application's read framebuffer, pack buffer and pack state are saved and put back, so
// the readback is invisible to it.
namespace GL = MG_Impl::GLImpl;
GLint readFramebuffer = 0, packBuffer = 0, alignment = 4, rowLength = 0, skipRows = 0, skipPixels = 0;
GL::GetIntegerv(GL_READ_FRAMEBUFFER_BINDING, &readFramebuffer);
GL::GetIntegerv(GL_PIXEL_PACK_BUFFER_BINDING, &packBuffer);
GL::GetIntegerv(GL_PACK_ALIGNMENT, &alignment);
GL::GetIntegerv(GL_PACK_ROW_LENGTH, &rowLength);
GL::GetIntegerv(GL_PACK_SKIP_ROWS, &skipRows);
GL::GetIntegerv(GL_PACK_SKIP_PIXELS, &skipPixels);
if (readFramebuffer != 0) GL::BindFramebuffer(GL_READ_FRAMEBUFFER, 0);
if (packBuffer != 0) GL::BindBuffer(GL_PIXEL_PACK_BUFFER, 0);
GL::PixelStorei(GL_PACK_ALIGNMENT, 4);
GL::PixelStorei(GL_PACK_ROW_LENGTH, 0);
GL::PixelStorei(GL_PACK_SKIP_ROWS, 0);
GL::PixelStorei(GL_PACK_SKIP_PIXELS, 0);
const SizeT rowBytes = static_cast<SizeT>(impl.width) * 4;
impl.scratch.resize(rowBytes * static_cast<SizeT>(impl.height));
GL::ReadPixels(0, 0, impl.width, impl.height, GL_RGBA, GL_UNSIGNED_BYTE, impl.scratch.data());
GL::PixelStorei(GL_PACK_ALIGNMENT, alignment);
GL::PixelStorei(GL_PACK_ROW_LENGTH, rowLength);
GL::PixelStorei(GL_PACK_SKIP_ROWS, skipRows);
GL::PixelStorei(GL_PACK_SKIP_PIXELS, skipPixels);
if (packBuffer != 0) GL::BindBuffer(GL_PIXEL_PACK_BUFFER, static_cast<GLuint>(packBuffer));
if (readFramebuffer != 0) GL::BindFramebuffer(GL_READ_FRAMEBUFFER, static_cast<GLuint>(readFramebuffer));
// GL's rows run bottom-up and RGBA; a wl_shm buffer's run top-down and B, G, R, A.
auto* dst = static_cast<Uint8*>(buffer->map);
for (EGLint y = 0; y < impl.height; ++y) {
const Uint8* src = impl.scratch.data() + static_cast<SizeT>(impl.height - 1 - y) * rowBytes;
Uint8* row = dst + static_cast<SizeT>(y) * static_cast<SizeT>(buffer->stride);
for (EGLint x = 0; x < impl.width; ++x) {
row[x * 4 + 0] = src[x * 4 + 2];
row[x * 4 + 1] = src[x * 4 + 1];
row[x * 4 + 2] = src[x * 4 + 0];
row[x * 4 + 3] = src[x * 4 + 3];
}
}
wl_proxy* surface = impl.window->surface;
const uint32_t surfaceVersion = api.getVersion(surface);
api.marshalFlags(surface, kSurfaceAttach, nullptr, surfaceVersion, 0, buffer->buffer, impl.window->dx,
impl.window->dy);
if (surfaceVersion >= 4) {
api.marshalFlags(surface, kSurfaceDamageBuffer, nullptr, surfaceVersion, 0, 0, 0, impl.width,
impl.height);
} else {
api.marshalFlags(surface, kSurfaceDamage, nullptr, surfaceVersion, 0, 0, 0, impl.width, impl.height);
}
api.marshalFlags(surface, kSurfaceCommit, nullptr, surfaceVersion, 0);
buffer->busy = true;
impl.window->attached_width = impl.width;
impl.window->attached_height = impl.height;
impl.window->dx = 0;
impl.window->dy = 0;
api.flush(impl.display);
return true;
}
} // namespace MobileGL::MG_Impl::EGLImpl::Wayland
#endif
+72
View File
@@ -0,0 +1,72 @@
// MobileGL - MobileGL/MG_Impl/EGLImpl/WaylandWindow.h
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#pragma once
// WAYLAND WINDOW SURFACES. A Wayland client hands eglCreateWindowSurface a wl_egl_window, and what
// the compositor shows is whatever wl_buffer the client attaches to that window's wl_surface. The
// pixels of a MobileGL frame are drawn where the backend is - on a split build, in another process
// on another OS - so a window surface here is a pbuffer the backend draws into, plus this: at every
// swap the frame is read back and attached to the wl_surface as a wl_shm buffer.
//
// libwayland-client is loaded at run time (it is already in every Wayland client's process), so the
// library neither links it nor needs its headers to build.
#include <Includes.h>
#if defined(__linux__) && !defined(__ANDROID__)
#define MOBILEGL_WAYLAND_WINDOWS 1
#else
#define MOBILEGL_WAYLAND_WINDOWS 0
#endif
#if MOBILEGL_WAYLAND_WINDOWS
#include <EGL/egl.h>
namespace MobileGL::MG_Impl::EGLImpl::Wayland {
// EGL_PLATFORM_WAYLAND_KHR and EGL_PLATFORM_WAYLAND_EXT are the same enum.
inline constexpr EGLenum kPlatformWayland = 0x31D8;
// Whether a display is a Wayland one: asked for as EGL_PLATFORM_WAYLAND, or (eglGetDisplay with
// no platform) a native display that is a wl_display - recognised the way Mesa does, by its
// first word being libwayland-client's wl_display_interface.
Bool IsWaylandDisplay(Uint64 nativeDisplay, EGLenum platform);
// Whether `window` looks like a wl_egl_window this library can present to.
Bool IsWaylandWindow(const void* window);
// The size the application last gave its wl_egl_window.
Bool WindowSize(const void* window, EGLint* width, EGLint* height);
// One window surface's presentation: the shm buffers, the event queue they are dispatched on,
// and the readback scratch. Created for a wl_egl_window on a wl_display; null with the reason
// logged when libwayland-client or the compositor's wl_shm is not there.
class WindowSurface {
public:
static UniquePtr<WindowSurface> Create(void* wlDisplay, void* wlEglWindow, EGLint width, EGLint height,
Bool hasAlpha);
~WindowSurface();
WindowSurface(const WindowSurface&) = delete;
WindowSurface& operator=(const WindowSurface&) = delete;
// Reads the frame back out of the current context's default framebuffer and puts it on
// the window: attach, damage, commit. The application's read-framebuffer and pack state
// are put back exactly as they were. False (logged) when the frame did not reach the
// window; the swap itself goes on regardless.
Bool Present();
struct Impl;
private:
explicit WindowSurface(UniquePtr<Impl> impl);
UniquePtr<Impl> m_impl;
};
} // namespace MobileGL::MG_Impl::EGLImpl::Wayland
#endif
+481 -12
View File
@@ -6,6 +6,9 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// The dma-buf import's client half: eglCreateImageKHR keeps a dup() of every plane's descriptor,
// and this is what tells the render server which texture that buffer backs (see the header).
#include <MG_Remote/Client/DmaBufImport.h>
#include <Includes.h>
#include "../Buffer/GL_Buffer.h"
#include "../Getter/GL_Getter.h"
@@ -22,6 +25,13 @@
#include "../Sync/GL_Sync.h"
#include "../Debug/GL_Debug.h"
#include <MG_State/GLState/Core.h>
// The EGL side of the state, for the two GL_OES_EGL_image entry points: what an EGLImage handle
// names is an object eglCreateImageKHR built (its geometry, its fourcc and the descriptors it owns),
// and that object lives in the EGL state. Same include the other GLImpl files that touch it use.
#include <MG_State/EGLState/Core.h>
// The compositor's frame boundary: glFlush presents the display host's frame when the surfaceless
// bind hook took one (see EGLImpl.cpp). The EGL side declares the pair and the swap entry.
#include <MG_Impl/EGLImpl/EGLImpl.h>
#define DECLARE_GL_FUNCTION_STUB_HEAD(type, name, ...) MOBILEGL_GL_API type gl##name(__VA_ARGS__) {
@@ -120,7 +130,23 @@ MOBILEGL_GL_API void glFinish() {
MobileGL::MG_Pipe::MGPipeClientFinish();
#endif
}
MOBILEGL_GL_API void glFlush() { MGLOG_D("Implementing function: %s(...)", __FUNCTION__); }
MOBILEGL_GL_API void glFlush() {
MGLOG_D("Implementing function: %s(...)", __FUNCTION__);
#if MOBILEGL_BUILD_DISAGGREGATED
MobileGL::MG_Pipe::MGPipeClientFlush();
#endif
// A COMPOSITOR'S FRAME BOUNDARY. kwin binds the display host's frame surface and draws the
// desktop into it, then calls glFlush - and glFlush alone. It never calls eglSwapBuffers, so
// without this the stream reached the server and the server drew, but the display host was
// never told a frame was done: the host's own present loop (AHardwareBuffer_acquire ->
// draw -> notify) is what puts pixels on the glass, and its notify is exactly this swap.
// Guarded by whether the surfaceless-bind hook actually took the host frame.
if (MobileGL::MG_Impl::EGLImpl::g_hostFramePresentSurface != nullptr &&
MobileGL::MG_Impl::EGLImpl::g_hostFramePresentDisplay != nullptr) {
MobileGL::MG_Impl::EGLImpl::SwapBuffers(MobileGL::MG_Impl::EGLImpl::g_hostFramePresentDisplay,
MobileGL::MG_Impl::EGLImpl::g_hostFramePresentSurface);
}
}
DECLARE_GL_FUNCTION_HEAD(void, FramebufferRenderbuffer, GLenum target, GLenum attachment, GLenum renderbuffertarget, GLuint renderbuffer) DECLARE_GL_FUNCTION_END_NO_RETURN(void, FramebufferRenderbuffer, target, attachment, renderbuffertarget, renderbuffer)
DECLARE_GL_FUNCTION_HEAD(void, FramebufferTexture2D, GLenum target, GLenum attachment, GLenum textarget, GLuint texture, GLint level) DECLARE_GL_FUNCTION_END_NO_RETURN(void, FramebufferTexture2D, target, attachment, textarget, texture, level)
DECLARE_GL_FUNCTION_HEAD(void, FrontFace, GLenum mode) DECLARE_GL_FUNCTION_END_NO_RETURN(void, FrontFace, mode)
@@ -1292,17 +1318,325 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, GetnMinmaxARB, GLenum target, GLboolean rese
DECLARE_GL_FUNCTION_STUB_HEAD(void, FramebufferSampleLocationsfvARB, GLenum target, GLuint start, GLsizei count, const GLfloat* v) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, FramebufferSampleLocationsfvARB, target, start, count, v)
DECLARE_GL_FUNCTION_STUB_HEAD(void, NamedFramebufferSampleLocationsfvARB, GLuint framebuffer, GLuint start, GLsizei count, const GLfloat* v) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, NamedFramebufferSampleLocationsfvARB, framebuffer, start, count, v)
DECLARE_GL_FUNCTION_STUB_HEAD(void, EvaluateDepthValuesARB, void) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, EvaluateDepthValuesARB, )
DECLARE_GL_FUNCTION_STUB_HEAD(void, DeleteObjectARB, GLhandleARB obj) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DeleteObjectARB, obj)
DECLARE_GL_FUNCTION_STUB_HEAD(GLhandleARB, GetHandleARB, GLenum pname) DECLARE_GL_FUNCTION_STUB_END(GLhandleARB, GetHandleARB, pname)
DECLARE_GL_FUNCTION_STUB_HEAD(void, DetachObjectARB, GLhandleARB containerObj, GLhandleARB attachedObj) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DetachObjectARB, containerObj, attachedObj)
DECLARE_GL_FUNCTION_STUB_HEAD(GLhandleARB, CreateShaderObjectARB, GLenum shaderType) DECLARE_GL_FUNCTION_STUB_END(GLhandleARB, CreateShaderObjectARB, shaderType)
DECLARE_GL_FUNCTION_STUB_HEAD(GLhandleARB, CreateProgramObjectARB, void) DECLARE_GL_FUNCTION_STUB_END(GLhandleARB, CreateProgramObjectARB, )
DECLARE_GL_FUNCTION_STUB_HEAD(void, AttachObjectARB, GLhandleARB containerObj, GLhandleARB obj) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, AttachObjectARB, containerObj, obj)
DECLARE_GL_FUNCTION_STUB_HEAD(void, UseProgramObjectARB, GLhandleARB programObj) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, UseProgramObjectARB, programObj)
DECLARE_GL_FUNCTION_STUB_HEAD(void, GetObjectParameterfvARB, GLhandleARB obj, GLenum pname, GLfloat* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetObjectParameterfvARB, obj, pname, params)
DECLARE_GL_FUNCTION_STUB_HEAD(void, GetObjectParameterivARB, GLhandleARB obj, GLenum pname, GLint* params) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetObjectParameterivARB, obj, pname, params)
DECLARE_GL_FUNCTION_STUB_HEAD(void, GetInfoLogARB, GLhandleARB obj, GLsizei maxLength, GLsizei* length, GLcharARB* infoLog) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetInfoLogARB, obj, maxLength, length, infoLog)
DECLARE_GL_FUNCTION_STUB_HEAD(void, GetAttachedObjectsARB, GLhandleARB containerObj, GLsizei maxCount, GLsizei* count, GLhandleARB* obj) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetAttachedObjectsARB, containerObj, maxCount, count, obj)
// GL_ARB_shader_objects, GL_ARB_vertex_shader, GL_ARB_fragment_shader.
//
// kwin_wayland 6.7.3 refuses a MobileGL context over three of the names these extensions add.
// EglContext::checkSupported() (src/opengl/eglcontext.cpp) takes the DESKTOP-OpenGL branch for
// every context whose GL_VERSION is not an "OpenGL ES" string:
// const bool supportsGLSL = m_isOpenglES || (hasOpenglExtension("GL_ARB_shader_objects") &&
// hasOpenglExtension("GL_ARB_fragment_shader") && hasOpenglExtension("GL_ARB_vertex_shader"));
// GL_VERSION here is "4.6.0 MobileGL ...", so all three names were REQUIRED of a context that did
// not carry them: checkSupported() answered false, EglContext::create() returned nullptr,
// EglBackend::createContext() returned false, and kwin printed "Could not initialize rendering
// context" and quit without drawing a frame. Both backends now advertise the three, and this block
// is what makes that advertisement true instead of a test unlock - every entry point the three
// extensions define is implemented here, over the shader and program objects the core entry points
// already own:
//
// ARB_shader_objects the 39 functions of the block include/GL/glext.h:4095-4173 declares.
// ARB_vertex_shader the three attrib-location entry points of glext.h:4856-4862.
// ARB_fragment_shader none. Its "New Procedures and Functions" section reads "None" - it is the
// capability half of the pair, and the reason kwin names it at all.
//
// HANDLES. The ARB family names objects with GLhandleARB. This repository's own glext.h types that
// as void* on Apple and as unsigned int on every other platform (glext.h:4055-4059, the same split
// the system headers carry), so on Linux, Windows and Android a handle IS the GLuint name this
// implementation hands out and both conversions below are the identity there. The pointer arm is
// not decoration: a caller on Apple passes a pointer, and an id written straight into its void*
// slot would be a size error rather than a cast error. H is a template parameter, and both arms
// cast through it, so the arm a platform does not use is never instantiated - and never has to be
// well-formed for the other typedef: reinterpret_cast between two integers, which is what the
// pointer arm becomes on Linux, is not a cast clang accepts. The shape is the one
// MG_State/EGLState/Core.h already uses for EGL's handles.
template <typename H = GLhandleARB>
static GLhandleARB ToGLHandleARB(GLuint name) {
if constexpr (std::is_pointer_v<H>) {
return reinterpret_cast<H>(static_cast<std::uintptr_t>(name));
} else {
return static_cast<H>(name);
}
}
template <typename H = GLhandleARB>
static GLuint FromGLHandleARB(GLhandleARB handle) {
if constexpr (std::is_pointer_v<H>) {
return static_cast<GLuint>(reinterpret_cast<std::uintptr_t>(static_cast<H>(handle)));
} else {
return static_cast<GLuint>(handle);
}
}
// The answer behind GetObjectParameter{if}vARB, and the reason neither of them carries a pname
// translation table: the GL_OBJECT_*_ARB tokens ARE the core tokens, numerically. Core GL took the
// object pnames over unchanged when ARB_shader_objects was folded in -
// GL_OBJECT_DELETE_STATUS_ARB through GL_OBJECT_SHADER_SOURCE_LENGTH_ARB are 0x8B80-0x8B88, the
// same numbers glcorearb.h gives GL_DELETE_STATUS through GL_SHADER_SOURCE_LENGTH
// (glext.h:4086-4094) - so the mapping is a RENAME and the same pname goes to the core getter of
// the same object kind. Handing it to that getter is also the right ERROR behaviour, and the reason
// this does not whitelist pnames: GL_OBJECT_LINK_STATUS_ARB asked of a SHADER has to come back
// INVALID_ENUM, and glGetShaderiv is where the set of pnames a shader accepts is defined.
//
// The one token core never took is GL_OBJECT_TYPE_ARB (0x8B4E; core's GL_OBJECT_TYPE, 0x9112, is
// ARB_program_interface_query's interface class, a different question with a different value
// space), so the object kind is the single thing answered here without a core getter. It is also
// the only way an ARB-era caller can ask whether a handle names a program or a shader: there is no
// glIsProgramARB/glIsShaderARB in ARB_shader_objects (include/GL/glext.h:4095-4173 declares
// neither, and the glIsProgramARB the registry does know, glext.h:3280, belongs to
// ARB_vertex_program/ARB_fragment_program - those ASM program objects are not these).
static MobileGL::Bool AnswerObjectParameterARB(GLuint name, GLenum pname, GLint* params) {
if (pname == GL_OBJECT_TYPE_ARB) {
if (glIsProgram(name)) {
*params = GL_PROGRAM_OBJECT_ARB;
return true;
}
if (glIsShader(name)) {
*params = GL_SHADER_OBJECT_ARB;
return true;
}
// The error ARB_vertex_shader's own text specifies: "The error INVALID_VALUE is generated
// by any command that takes one or more handles as input, and one or more handles are not
// an object handle generated by OpenGL."
MobileGL::MG_State::pGLContext->RecordError(
MobileGL::ErrorCode::InvalidValue,
MobileGL::MakeUnique<MobileGL::GenericErrorInfo>(
"MG_Impl/GLImpl", __func__,
std::to_string(name) + " is not an object handle generated by OpenGL."));
return false;
}
if (glIsProgram(name)) {
glGetProgramiv(name, pname, params);
return true;
}
// A shader, or a name that is neither. glGetShaderiv owns both the shader pname set and the
// name validation, which is the one place that splits a program name handed to a shader query
// (INVALID_OPERATION) from a name GL never generated (INVALID_VALUE) - the split GL 3.3 core
// 2.11 requires. It writes nothing in either error case, which is what the caller reads as
// "no answer".
const MobileGL::Bool isShader = glIsShader(name) == GL_TRUE;
glGetShaderiv(name, pname, params);
return isShader;
}
MOBILEGL_GL_API void glDeleteObjectARB(GLhandleARB obj) {
// ARB_shader_objects has ONE object namespace - a handle names a program or a shader and
// nothing else - and DeleteObjectARB is the deletion entry point for both kinds, so the kind
// is decided here. IsProgram/IsShader answer over the same single namespace this
// implementation hands names out of, so a program name cannot be taken for a shader name.
// Zero, and a handle that names neither object, are ignored: there is no object to delete.
const GLuint name = FromGLHandleARB(obj);
if (glIsProgram(name)) {
glDeleteProgram(name);
} else if (glIsShader(name)) {
glDeleteShader(name);
}
}
MOBILEGL_GL_API GLhandleARB glGetHandleARB(GLenum pname) {
if (pname != GL_PROGRAM_OBJECT_ARB) {
MobileGL::MG_State::pGLContext->RecordError(
MobileGL::ErrorCode::InvalidEnum,
MobileGL::MakeUnique<MobileGL::GenericErrorInfo>(
"MG_Impl/GLImpl", __func__,
"pname " + std::to_string(pname) + " is not GL_PROGRAM_OBJECT_ARB."));
return ToGLHandleARB(0u);
}
// "The program object currently in use", read from the state GL_CURRENT_PROGRAM is answered
// from - so a program bound through glUseProgram is found here too, and a
// UseProgramObjectARB(0) reads back as the zero handle.
const auto& currentProgram = MobileGL::MG_State::pGLContext->GetCurrentProgram();
return ToGLHandleARB(currentProgram ? static_cast<GLuint>(currentProgram->GetExternalIndex()) : 0u);
}
MOBILEGL_GL_API void glDetachObjectARB(GLhandleARB containerObj, GLhandleARB attachedObj) {
glDetachShader(FromGLHandleARB(containerObj), FromGLHandleARB(attachedObj));
}
MOBILEGL_GL_API GLhandleARB glCreateShaderObjectARB(GLenum shaderType) {
return ToGLHandleARB(glCreateShader(shaderType));
}
MOBILEGL_GL_API void glShaderSourceARB(GLhandleARB shaderObj, GLsizei count, const GLcharARB** string,
const GLint* length) {
glShaderSource(FromGLHandleARB(shaderObj), count, string, length);
}
MOBILEGL_GL_API void glCompileShaderARB(GLhandleARB shaderObj) {
glCompileShader(FromGLHandleARB(shaderObj));
}
MOBILEGL_GL_API GLhandleARB glCreateProgramObjectARB(void) {
return ToGLHandleARB(glCreateProgram());
}
MOBILEGL_GL_API void glAttachObjectARB(GLhandleARB containerObj, GLhandleARB obj) {
glAttachShader(FromGLHandleARB(containerObj), FromGLHandleARB(obj));
}
MOBILEGL_GL_API void glLinkProgramARB(GLhandleARB programObj) {
glLinkProgram(FromGLHandleARB(programObj));
}
MOBILEGL_GL_API void glUseProgramObjectARB(GLhandleARB programObj) {
glUseProgram(FromGLHandleARB(programObj));
}
MOBILEGL_GL_API void glValidateProgramARB(GLhandleARB programObj) {
glValidateProgram(FromGLHandleARB(programObj));
}
MOBILEGL_GL_API void glUniform1fARB(GLint location, GLfloat v0) {
glUniform1f(location, v0);
}
MOBILEGL_GL_API void glUniform2fARB(GLint location, GLfloat v0, GLfloat v1) {
glUniform2f(location, v0, v1);
}
MOBILEGL_GL_API void glUniform3fARB(GLint location, GLfloat v0, GLfloat v1, GLfloat v2) {
glUniform3f(location, v0, v1, v2);
}
MOBILEGL_GL_API void glUniform4fARB(GLint location, GLfloat v0, GLfloat v1, GLfloat v2, GLfloat v3) {
glUniform4f(location, v0, v1, v2, v3);
}
MOBILEGL_GL_API void glUniform1iARB(GLint location, GLint v0) {
glUniform1i(location, v0);
}
MOBILEGL_GL_API void glUniform2iARB(GLint location, GLint v0, GLint v1) {
glUniform2i(location, v0, v1);
}
MOBILEGL_GL_API void glUniform3iARB(GLint location, GLint v0, GLint v1, GLint v2) {
glUniform3i(location, v0, v1, v2);
}
MOBILEGL_GL_API void glUniform4iARB(GLint location, GLint v0, GLint v1, GLint v2, GLint v3) {
glUniform4i(location, v0, v1, v2, v3);
}
MOBILEGL_GL_API void glUniform1fvARB(GLint location, GLsizei count, const GLfloat* value) {
glUniform1fv(location, count, value);
}
MOBILEGL_GL_API void glUniform2fvARB(GLint location, GLsizei count, const GLfloat* value) {
glUniform2fv(location, count, value);
}
MOBILEGL_GL_API void glUniform3fvARB(GLint location, GLsizei count, const GLfloat* value) {
glUniform3fv(location, count, value);
}
MOBILEGL_GL_API void glUniform4fvARB(GLint location, GLsizei count, const GLfloat* value) {
glUniform4fv(location, count, value);
}
MOBILEGL_GL_API void glUniform1ivARB(GLint location, GLsizei count, const GLint* value) {
glUniform1iv(location, count, value);
}
MOBILEGL_GL_API void glUniform2ivARB(GLint location, GLsizei count, const GLint* value) {
glUniform2iv(location, count, value);
}
MOBILEGL_GL_API void glUniform3ivARB(GLint location, GLsizei count, const GLint* value) {
glUniform3iv(location, count, value);
}
MOBILEGL_GL_API void glUniform4ivARB(GLint location, GLsizei count, const GLint* value) {
glUniform4iv(location, count, value);
}
MOBILEGL_GL_API void glUniformMatrix2fvARB(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value) {
glUniformMatrix2fv(location, count, transpose, value);
}
MOBILEGL_GL_API void glUniformMatrix3fvARB(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value) {
glUniformMatrix3fv(location, count, transpose, value);
}
MOBILEGL_GL_API void glUniformMatrix4fvARB(GLint location, GLsizei count, GLboolean transpose, const GLfloat* value) {
glUniformMatrix4fv(location, count, transpose, value);
}
MOBILEGL_GL_API void glGetObjectParameterfvARB(GLhandleARB obj, GLenum pname, GLfloat* params) {
// The fv form asks the same questions - the ARB object model holds no float-valued object
// state - so the answer is the iv answer widened. A query the iv path refuses leaves the
// output untouched here as well, rather than putting a plausible-looking 0.0 where there is
// no answer.
GLint value = 0;
if (!AnswerObjectParameterARB(FromGLHandleARB(obj), pname, &value)) return;
*params = static_cast<GLfloat>(value);
}
MOBILEGL_GL_API void glGetObjectParameterivARB(GLhandleARB obj, GLenum pname, GLint* params) {
AnswerObjectParameterARB(FromGLHandleARB(obj), pname, params);
}
MOBILEGL_GL_API void glGetInfoLogARB(GLhandleARB obj, GLsizei maxLength, GLsizei* length, GLcharARB* infoLog) {
// The same handle-kind dispatch as AnswerObjectParameterARB, and the same fallback: the shader
// getter is also where the name validation and its error code live.
const GLuint name = FromGLHandleARB(obj);
if (glIsProgram(name)) {
glGetProgramInfoLog(name, maxLength, length, infoLog);
} else {
glGetShaderInfoLog(name, maxLength, length, infoLog);
}
}
MOBILEGL_GL_API void glGetAttachedObjectsARB(GLhandleARB containerObj, GLsizei maxCount, GLsizei* count,
GLhandleARB* obj) {
// The caller's array holds GLhandleARB while glGetAttachedShaders writes GLuint names, and on
// Apple those are different sizes, so the ids are gathered first and converted one at a time.
// A negative maxCount is passed through: glGetAttachedShaders is where its INVALID_VALUE is.
MobileGL::Vector<GLuint> shaders(maxCount > 0 ? static_cast<MobileGL::SizeT>(maxCount) : 0, 0);
GLsizei attached = 0;
glGetAttachedShaders(FromGLHandleARB(containerObj), maxCount, &attached, shaders.data());
for (GLsizei i = 0; i < attached; ++i) {
obj[i] = ToGLHandleARB(shaders[i]);
}
if (count) *count = attached;
}
MOBILEGL_GL_API GLint glGetUniformLocationARB(GLhandleARB programObj, const GLcharARB* name) {
return glGetUniformLocation(FromGLHandleARB(programObj), name);
}
MOBILEGL_GL_API void glGetActiveUniformARB(GLhandleARB programObj, GLuint index, GLsizei maxLength,
GLsizei* length, GLint* size, GLenum* type, GLcharARB* name) {
glGetActiveUniform(FromGLHandleARB(programObj), index, maxLength, length, size, type, name);
}
MOBILEGL_GL_API void glGetUniformfvARB(GLhandleARB programObj, GLint location, GLfloat* params) {
glGetUniformfv(FromGLHandleARB(programObj), location, params);
}
MOBILEGL_GL_API void glGetUniformivARB(GLhandleARB programObj, GLint location, GLint* params) {
glGetUniformiv(FromGLHandleARB(programObj), location, params);
}
MOBILEGL_GL_API void glGetShaderSourceARB(GLhandleARB obj, GLsizei maxLength, GLsizei* length,
GLcharARB* source) {
glGetShaderSource(FromGLHandleARB(obj), maxLength, length, source);
}
// ARB_vertex_shader's three entry points. The generic-attribute family that extension also
// describes is spelled with the ARB_vertex_program names (glVertexAttribPointerARB and the rest),
// which glext.h declares in that other block and which MobileGL does not advertise; nothing here
// claims them.
MOBILEGL_GL_API void glBindAttribLocationARB(GLhandleARB programObj, GLuint index, const GLcharARB* name) {
glBindAttribLocation(FromGLHandleARB(programObj), index, name);
}
MOBILEGL_GL_API void glGetActiveAttribARB(GLhandleARB programObj, GLuint index, GLsizei maxLength,
GLsizei* length, GLint* size, GLenum* type, GLcharARB* name) {
glGetActiveAttrib(FromGLHandleARB(programObj), index, maxLength, length, size, type, name);
}
MOBILEGL_GL_API GLint glGetAttribLocationARB(GLhandleARB programObj, const GLcharARB* name) {
return glGetAttribLocation(FromGLHandleARB(programObj), name);
}
DECLARE_GL_FUNCTION_STUB_HEAD(void, NamedStringARB, GLenum type, GLint namelen, const GLchar* name, GLint strinen, const GLchar* string) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, NamedStringARB, type, namelen, name, strinen, string)
DECLARE_GL_FUNCTION_STUB_HEAD(void, DeleteNamedStringARB, GLint namelen, const GLchar* name) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, DeleteNamedStringARB, namelen, name)
DECLARE_GL_FUNCTION_STUB_HEAD(void, CompileShaderIncludeARB, GLuint shader, GLsizei count, const GLchar* const* path, const GLint* length) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompileShaderIncludeARB, shader, count, path, length)
@@ -1708,6 +2042,141 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, VertexBlendEnviATI, GLenum pname, GLint para
DECLARE_GL_FUNCTION_STUB_HEAD(void, VertexBlendEnvfATI, GLenum pname, GLfloat param) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, VertexBlendEnvfATI, pname, param)
DECLARE_GL_FUNCTION_STUB_HEAD(void, EGLImageTargetTexStorageEXT, GLenum target, GLeglImageOES image, const GLint* attrib_list) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, EGLImageTargetTexStorageEXT, target, image, attrib_list)
DECLARE_GL_FUNCTION_STUB_HEAD(void, EGLImageTargetTextureStorageEXT, GLuint texture, GLeglImageOES image, const GLint* attrib_list) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, EGLImageTargetTextureStorageEXT, texture, image, attrib_list)
// GL_OES_EGL_image.
//
// MEASURED, and the whole reason this block exists: libepoxy in kwin_wayland prints
// No provider of glEGLImageTargetTexture2DOES found. Requires one of: GL_OES_EGL_image
// and refuses the import path there - and that path is how the compositor's scene reaches the
// display daemon's dma-bufs, i.e. how a Linux desktop gets onto the phone's screen. The EGL half is
// real: eglCreateImageKHR builds an EGL_LINUX_DMA_BUF_EXT image that owns a dup() of every plane's
// descriptor and carries that buffer's geometry and fourcc (MG_State/EGLState/Core.h). So a handle
// that arrives here is resolvable, and what these two entry points do is exactly that: resolve it,
// refuse what names nothing, and say ONCE what is not wired yet. What is not wired yet is the
// render-server half - telling the server which texture a dma-buf backs, over the one channel that
// can carry a descriptor across the container boundary - so a framebuffer built over an imported
// texture is not complete until that lands. Pretending otherwise (attaching the image to a texture
// object that has no storage behind it) would turn "not implemented" into "mysteriously black",
// which is what the log line below is here to prevent.
MOBILEGL_GL_API void glEGLImageTargetTexture2DOES(GLenum target, GLeglImageOES image) {
// 0x8D65 is GL_TEXTURE_EXTERNAL_OES, which the headers this file compiles against do not spell
// out. It matters here MORE than GL_TEXTURE_2D does: a compositor that could not map an
// imported dma-buf's format to a non-external one binds its image to THIS target, so refusing it
// would turn "no format map yet" into an INVALID_ENUM the caller cannot act on (kwin 6.7.3 does
// exactly that when EglDisplay::queryImportFormats() answers nothing).
constexpr GLenum kTextureExternalOes = 0x8D65;
if (target != GL_TEXTURE_2D && target != kTextureExternalOes) {
MobileGL::MG_State::pGLContext->RecordError(
MobileGL::ErrorCode::InvalidEnum,
MobileGL::MakeUnique<MobileGL::GenericErrorInfo>("MG_Impl/GLImpl", __func__,
"target is not a texture target this entry point takes."));
return;
}
EGLint imageWidth = 0;
EGLint imageHeight = 0;
EGLint imageFourCC = 0;
EGLint imagePlanes = 0;
const MobileGL::Bool imageKnown =
MobileGL::MG_State::pEGLContext != nullptr &&
MobileGL::MG_State::pEGLContext->DescribeImage(reinterpret_cast<::EGLImage>(image), &imageWidth, &imageHeight,
&imageFourCC, &imagePlanes);
if (!imageKnown) {
// A handle this side did not hand out: ARB/OES text puts that on INVALID_VALUE, and it is
// the answer a caller can act on rather than a texture that renders nowhere.
MobileGL::MG_State::pGLContext->RecordError(
MobileGL::ErrorCode::InvalidValue,
MobileGL::MakeUnique<MobileGL::GenericErrorInfo>("MG_Impl/GLImpl", __func__,
"image is not an EGLImage this display created."));
return;
}
MGLOG_W_ONCE("glEGLImageTargetTexture2DOES: taking target 0x%04x over an EGLImage of %dx%d fourcc 0x%08x "
"(%d plane(s)) that this side holds; the render server has NOT been told which texture that "
"dma-buf backs yet, so a framebuffer over this texture is not complete (the fd transport over "
"the GBM socket and the server-side import are the next change)",
target, imageWidth, imageHeight, imageFourCC, imagePlanes);
// AND NOW WIRE IT: a texture bound to an imported EGLImage has no storage on the render server,
// and a framebuffer built over it there is not complete - kwin 6.7.3 aborts on its own
// RenderTarget assertion right after this log line (measured twice). So the server is told which
// buffer backs the texture: one message per plane, each carrying that plane's descriptor over the
// session transport's SCM_RIGHTS channel. No session, or a transport without that channel, is a
// refusal by name from ShareImportedDmaBuf - the texture simply keeps having no storage, which is
// the honest state rather than a silent render into nothing.
MGLOG_I("dma-buf import: entry (target=0x%04x, image=%p)", static_cast<unsigned>(target), static_cast<void*>(image));
if (MobileGL::MG_State::pEGLContext != nullptr) {
constexpr MobileGL::Int kDmaBufImportPlanes = 4;
EGLint planeCount = 0;
EGLint planeFds[kDmaBufImportPlanes] = {-1, -1, -1, -1};
EGLint planeOffsets[kDmaBufImportPlanes] = {0, 0, 0, 0};
EGLint planePitches[kDmaBufImportPlanes] = {0, 0, 0, 0};
MobileGL::Uint64 planeModifiers[kDmaBufImportPlanes] = {0, 0, 0, 0};
MobileGL::Bool planeHasModifier[kDmaBufImportPlanes] = {false, false, false, false};
if (MobileGL::MG_State::pEGLContext->DescribeImagePlanes(
reinterpret_cast<::EGLImage>(image), &planeCount, planeFds, planeOffsets, planePitches,
planeModifiers, planeHasModifier, kDmaBufImportPlanes)) {
// The texture the image was bound to. 0x8D65 has no TextureTarget member in this state
// layer (nothing else here binds it), so the name is looked up through Texture2D and may
// come back 0 - the server logs what it was told rather than guessing.
MobileGL::Uint32 texName = 0;
if (MobileGL::MG_State::pGLContext != nullptr) {
const MobileGL::Int unit = MobileGL::MG_State::pGLContext->GetActiveTextureUnit();
// THE EXTERNAL SLOT, which is where a compositor's imported dma-buf texture lives.
const auto& bound = MobileGL::MG_State::pGLContext->GetTextureUnitObject(unit)
.GetBindingSlot(MobileGL::TextureTarget::TextureExternalOES)
.GetBoundObject();
if (bound != nullptr) {
texName = static_cast<MobileGL::Uint32>(bound->GetExternalIndex());
// AN IMPORTED TEXTURE IS BACKED, SO IT IS COMPLETE. Without this the state layer
// answers incompletely (TextureObjectBase::IsComplete only looks at the internal
// format) and the compositor, having attached it to a framebuffer, reports
// "framebuffer for dmabuf 0 is not complete" and stops before drawing anything.
bound->SetInternalFormat(MobileGL::TextureInternalFormat::RGBA8);
}
}
MGLOG_I("dma-buf import: textures resolved (texName=%u planes=%d), sharing", texName,
static_cast<int>(planeCount));
for (MobileGL::Int plane = 0; plane < planeCount; plane++) {
if (planeFds[plane] < 0) continue;
MGLOG_I("dma-buf import: sharing plane %d fd=%d", static_cast<int>(plane), planeFds[plane]);
(void)MobileGL::MG_Remote::Client::ShareImportedDmaBuf(
static_cast<MobileGL::Uint32>(target), texName, static_cast<MobileGL::Uint32>(imageWidth),
static_cast<MobileGL::Uint32>(imageHeight), static_cast<MobileGL::Uint32>(imageFourCC),
static_cast<MobileGL::Uint32>(plane), static_cast<MobileGL::Uint32>(planeCount), planeFds[plane],
planePitches[plane], planeOffsets[plane], planeModifiers[plane], planeHasModifier[plane]);
MGLOG_I("dma-buf import: plane %d shared", static_cast<int>(plane));
}
}
}
}
MOBILEGL_GL_API void glEGLImageTargetRenderbufferStorageOES(GLenum target, GLeglImageOES image) {
if (target != GL_RENDERBUFFER) {
MobileGL::MG_State::pGLContext->RecordError(
MobileGL::ErrorCode::InvalidEnum,
MobileGL::MakeUnique<MobileGL::GenericErrorInfo>("MG_Impl/GLImpl", __func__,
"target is not GL_RENDERBUFFER."));
return;
}
EGLint imageWidth = 0;
EGLint imageHeight = 0;
EGLint imageFourCC = 0;
EGLint imagePlanes = 0;
const MobileGL::Bool imageKnown =
MobileGL::MG_State::pEGLContext != nullptr &&
MobileGL::MG_State::pEGLContext->DescribeImage(reinterpret_cast<::EGLImage>(image), &imageWidth, &imageHeight,
&imageFourCC, &imagePlanes);
if (!imageKnown) {
MobileGL::MG_State::pGLContext->RecordError(
MobileGL::ErrorCode::InvalidValue,
MobileGL::MakeUnique<MobileGL::GenericErrorInfo>("MG_Impl/GLImpl", __func__,
"image is not an EGLImage this display created."));
return;
}
MGLOG_W_ONCE("glEGLImageTargetRenderbufferStorageOES: the renderbuffer takes an EGLImage of %dx%d fourcc "
"0x%08x; the render-server half of the import is the next change, exactly as for the texture "
"entry point beside this one",
imageWidth, imageHeight, imageFourCC);
}
DECLARE_GL_FUNCTION_STUB_HEAD(void, UniformBufferEXT, GLuint program, GLint location, GLuint buffer) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, UniformBufferEXT, program, location, buffer)
DECLARE_GL_FUNCTION_STUB_HEAD(GLint, GetUniformBufferSizeEXT, GLuint program, GLint location) DECLARE_GL_FUNCTION_STUB_END(GLint, GetUniformBufferSizeEXT, program, location)
DECLARE_GL_FUNCTION_STUB_HEAD(GLintptr, GetUniformOffsetEXT, GLuint program, GLint location) DECLARE_GL_FUNCTION_STUB_END(GLintptr, GetUniformOffsetEXT, program, location)
@@ -100,6 +100,13 @@ namespace MobileGL::MG_Impl::GLImpl {
}
Bool HasDefinedAttachment(const MG_State::GLState::FramebufferObject& framebufferObject) {
// NOT gated on the host frame, and that is deliberate: this only decides the ERROR CODE
// once completeness has already failed (CheckFramebufferStatus_State asks
// CheckCompleteness first), and the imported dma-buf's attachment is counted as valid
// THERE (MG_State/GLState/FramebufferState/FramebufferObject.cpp, whose gate lands before
// its own validity test for exactly this reason). Touching it here would add a second
// place that has to know the backend's frame - and the frontend knowing that fact is the
// compromise the completeness comment already names the end of.
for (const auto& attachment : framebufferObject.GetAllAttachmentObjects()) {
if (attachment.IsValid() && !attachment.IsEmpty()) {
return true;
@@ -611,7 +618,9 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto expectedTextureTarget = MG_Util::ConvertTextureUploadTargetToTextureTarget(textureUploadTarget);
if (expectedTextureTarget == TextureTarget::Unknown ||
textureObject->GetTarget() != expectedTextureTarget) {
(textureObject->GetTarget() != expectedTextureTarget &&
!(textureObject->GetTarget() == TextureTarget::TextureExternalOES &&
expectedTextureTarget == TextureTarget::Texture2D))) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidOperation,
MakeUnique<GenericErrorInfo>(
@@ -1382,7 +1391,9 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto expectedTextureTarget = MG_Util::ConvertTextureUploadTargetToTextureTarget(textureUploadTarget);
if (expectedTextureTarget == TextureTarget::Unknown ||
textureObject->GetTarget() != expectedTextureTarget) {
(textureObject->GetTarget() != expectedTextureTarget &&
!(textureObject->GetTarget() == TextureTarget::TextureExternalOES &&
expectedTextureTarget == TextureTarget::Texture2D))) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidOperation,
MakeUnique<GenericErrorInfo>(
@@ -1532,7 +1543,9 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto expectedTextureTarget = MG_Util::ConvertTextureUploadTargetToTextureTarget(textureUploadTarget);
if (expectedTextureTarget == TextureTarget::Unknown ||
textureObject->GetTarget() != expectedTextureTarget) {
(textureObject->GetTarget() != expectedTextureTarget &&
!(textureObject->GetTarget() == TextureTarget::TextureExternalOES &&
expectedTextureTarget == TextureTarget::Texture2D))) {
MG_State::pGLContext->RecordError(
ErrorCode::InvalidOperation,
MakeUnique<GenericErrorInfo>(
@@ -41,6 +41,18 @@
// call sites below are declarations only, exactly as the frontend's mutators are.
#include <MG_Pipe/PipeMutation.h>
// P9 (W1, MG_Remote/CONTRACT-P9.md §1): AN UNPACK SOURCE IS READ FROM THE BUFFER'S CLIENT SHADOW,
// and under a transport a GPU write into that buffer - a read_pixels_to_buffer, which is how a
// glReadPixels into a pack buffer now crosses, or a shader store - reaches the shadow only through
// SyncGpuWrites, exactly as glMapBufferRange and glGetBufferSubData already do. Monolith keeps its
// read as it was: its ReadPixels maps the driver PBO back into the shadow inside the call, and the
// pull build compiles this to nothing (G1).
#if MOBILEGL_BUILD_DISAGGREGATED
#define MGL_SYNC_PIXEL_UNPACK_SOURCE(buffer) do { if (MG_Config::Transport != MG_Config::TransportMode::Monolith) (buffer)->SyncGpuWrites(); } while (0)
#else
#define MGL_SYNC_PIXEL_UNPACK_SOURCE(buffer) ((void)0)
#endif
namespace MobileGL::MG_Impl::GLImpl {
static SharedPtr<MG_State::GLState::ITextureObject> nullTextureObject;
static UnorderedMap<Uint, Bool> g_autoGenerateMipmapByTextureId;
@@ -1773,6 +1785,7 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto& unpackBuffer =
MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::PixelUnpack).GetBoundObject();
if (!unpackBuffer) return data;
MGL_SYNC_PIXEL_UNPACK_SOURCE(unpackBuffer);
return reinterpret_cast<const char*>(unpackBuffer->MappedData()) + reinterpret_cast<SizeT>(data);
}
@@ -1819,6 +1832,7 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto& pixelUnpackBufferObject =
MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::PixelUnpack).GetBoundObject();
if (pixelUnpackBufferObject) {
MGL_SYNC_PIXEL_UNPACK_SOURCE(pixelUnpackBufferObject);
originalPixels = reinterpret_cast<const char*>(pixelUnpackBufferObject->MappedData()) +
reinterpret_cast<SizeT>(pixels);
}
@@ -1941,6 +1955,7 @@ namespace MobileGL::MG_Impl::GLImpl {
if (pixelUnpackBufferObject) {
MGLOG_D("TexSubImage2D_State: Using Pixel Unpack Buffer Object ID: %u",
pixelUnpackBufferObject->GetExternalIndex());
MGL_SYNC_PIXEL_UNPACK_SOURCE(pixelUnpackBufferObject);
originalPixels = reinterpret_cast<const char*>(pixelUnpackBufferObject->MappedData()) +
reinterpret_cast<SizeT>(pixels);
}
@@ -2030,6 +2045,7 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto& pixelUnpackBufferObject =
MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::PixelUnpack).GetBoundObject();
if (pixelUnpackBufferObject) {
MGL_SYNC_PIXEL_UNPACK_SOURCE(pixelUnpackBufferObject);
originalPixels = reinterpret_cast<const char*>(pixelUnpackBufferObject->MappedData()) +
reinterpret_cast<SizeT>(pixels);
}
@@ -2535,6 +2551,7 @@ namespace MobileGL::MG_Impl::GLImpl {
if (pixelUnpackBufferObject) {
MGLOG_D("%s: Using Pixel Unpack Buffer Object ID: %u", __func__,
pixelUnpackBufferObject->GetExternalIndex());
MGL_SYNC_PIXEL_UNPACK_SOURCE(pixelUnpackBufferObject);
originalPixels = reinterpret_cast<const char*>(pixelUnpackBufferObject->MappedData()) +
reinterpret_cast<SizeT>(pixels);
}
@@ -2682,6 +2699,7 @@ namespace MobileGL::MG_Impl::GLImpl {
if (pixelUnpackBufferObject) {
MGLOG_D("%s: Using Pixel Unpack Buffer Object ID: %u", __func__,
pixelUnpackBufferObject->GetExternalIndex());
MGL_SYNC_PIXEL_UNPACK_SOURCE(pixelUnpackBufferObject);
originalPixels = reinterpret_cast<const char*>(pixelUnpackBufferObject->MappedData()) +
reinterpret_cast<SizeT>(pixels);
}
@@ -2800,6 +2818,7 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto& pixelUnpackBufferObject =
MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::PixelUnpack).GetBoundObject();
if (pixelUnpackBufferObject) {
MGL_SYNC_PIXEL_UNPACK_SOURCE(pixelUnpackBufferObject);
originalPixels = reinterpret_cast<const char*>(pixelUnpackBufferObject->MappedData()) +
reinterpret_cast<SizeT>(pixels);
}
@@ -6262,6 +6281,7 @@ namespace MobileGL::MG_Impl::GLImpl {
const auto& pixelUnpackBufferObject =
MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::PixelUnpack).GetBoundObject();
if (pixelUnpackBufferObject) {
MGL_SYNC_PIXEL_UNPACK_SOURCE(pixelUnpackBufferObject);
originalPixels = reinterpret_cast<const char*>(pixelUnpackBufferObject->MappedData()) +
reinterpret_cast<SizeT>(pixels);
}
+1 -1
View File
@@ -174,7 +174,7 @@ namespace MobileGL::MG_Impl::GLXImpl {
}
struct ContextObject {
Display* XDisplay = nullptr;
::Display* XDisplay = nullptr;
EGLDisplay Display = EGL_NO_DISPLAY;
EGLConfig Config = nullptr;
EGLContext Context = EGL_NO_CONTEXT;
+51
View File
@@ -869,6 +869,55 @@ namespace MobileGL::MG_Impl {
GETPROC(glNamedBufferPageCommitmentARB, name);
GETPROC(glTexPageCommitmentARB, name);
GETPROC(glTexBufferARB, name);
// ARB_shader_objects and ARB_vertex_shader, by name, because a name is the only way an
// application ever asks for an extension entry point. Measured path: kwin_wayland resolves GL
// through libepoxy, and libepoxy asks eglGetProcAddress -> EGLImpl::GetProcAddress -> this
// table. Before these lines the callers that need the extension whose STRINGS make
// checkSupported() pass got a nullptr for every function of it - dlsym on the library found
// them, this table did not. The strings and the implementations are advertised from
// MG_Impl/GLImpl/Exporting/Definitions.cpp and the two BackendObject_* extension lists.
GETPROC(glDeleteObjectARB, name);
GETPROC(glGetHandleARB, name);
GETPROC(glDetachObjectARB, name);
GETPROC(glCreateShaderObjectARB, name);
GETPROC(glShaderSourceARB, name);
GETPROC(glCompileShaderARB, name);
GETPROC(glCreateProgramObjectARB, name);
GETPROC(glAttachObjectARB, name);
GETPROC(glLinkProgramARB, name);
GETPROC(glUseProgramObjectARB, name);
GETPROC(glValidateProgramARB, name);
GETPROC(glUniform1fARB, name);
GETPROC(glUniform2fARB, name);
GETPROC(glUniform3fARB, name);
GETPROC(glUniform4fARB, name);
GETPROC(glUniform1iARB, name);
GETPROC(glUniform2iARB, name);
GETPROC(glUniform3iARB, name);
GETPROC(glUniform4iARB, name);
GETPROC(glUniform1fvARB, name);
GETPROC(glUniform2fvARB, name);
GETPROC(glUniform3fvARB, name);
GETPROC(glUniform4fvARB, name);
GETPROC(glUniform1ivARB, name);
GETPROC(glUniform2ivARB, name);
GETPROC(glUniform3ivARB, name);
GETPROC(glUniform4ivARB, name);
GETPROC(glUniformMatrix2fvARB, name);
GETPROC(glUniformMatrix3fvARB, name);
GETPROC(glUniformMatrix4fvARB, name);
GETPROC(glGetObjectParameterfvARB, name);
GETPROC(glGetObjectParameterivARB, name);
GETPROC(glGetInfoLogARB, name);
GETPROC(glGetAttachedObjectsARB, name);
GETPROC(glGetUniformLocationARB, name);
GETPROC(glGetActiveUniformARB, name);
GETPROC(glGetUniformfvARB, name);
GETPROC(glGetUniformivARB, name);
GETPROC(glGetShaderSourceARB, name);
GETPROC(glBindAttribLocationARB, name);
GETPROC(glGetActiveAttribARB, name);
GETPROC(glGetAttribLocationARB, name);
GETPROC(glDepthRangeArraydvNV, name);
GETPROC(glDepthRangeIndexeddNV, name);
GETPROC(glBlendBarrierKHR, name);
@@ -886,6 +935,8 @@ namespace MobileGL::MG_Impl {
GETPROC(glBeginPerfMonitorAMD, name);
GETPROC(glEndPerfMonitorAMD, name);
GETPROC(glGetPerfMonitorCounterDataAMD, name);
GETPROC(glEGLImageTargetTexture2DOES, name);
GETPROC(glEGLImageTargetRenderbufferStorageOES, name);
GETPROC(glEGLImageTargetTexStorageEXT, name);
GETPROC(glEGLImageTargetTextureStorageEXT, name);
GETPROC(glLabelObjectEXT, name);
+227 -81
View File
@@ -582,12 +582,25 @@ namespace MobileGL::MG_Pipe {
Optional<MGPipeInputField> Corrupt;
String CorruptKnob; // the MOBILEGL_PIPE_VERIFY_CORRUPT value the last arm saw
std::atomic<Uint64> Divergences{0};
~VerifyState() {
Uint64 Summarised = 0; // how many of them the summary line has reported so far
// THE SUMMARY IS WRITTEN FROM MobileGL::Destroy, NOT FROM HERE (V1 fix round 2). This
// is a namespace-scope static, so this destructor runs from __run_exit_handlers, AFTER
// DestroyImpl has called MG_Util::Debug::Close(). Close nulls the role's sink; Log.cpp's
// WriteToFile then re-runs InitFile() for the next line, and InitFile opens the role's
// file with "w". So the one line this destructor wrote was the only line the client half
// of every FATAL=0 run kept - 121 bytes, the summary itself, with the entry compare's
// own reports, the arming line and the config dump gone (and the monolith arm's
// VerifyCorrupted. log wiped the same way). MGPipeVerifyFlushSummary runs before Close,
// so by the time this destructor runs there is nothing new to say and it says nothing;
// the arm is kept for a process that never called Destroy, where no Close ran and the
// line lands in the still-open file as it always did.
~VerifyState() { FlushSummary(); }
void FlushSummary() {
const Uint64 count = Divergences.load(std::memory_order_relaxed);
if (count != 0) {
MGLOG_E("MGPipe: verify summary - %llu divergence(s) survived MOBILEGL_PIPE_VERIFY_FATAL=0",
static_cast<unsigned long long>(count));
}
if (count == 0 || count == Summarised) return;
Summarised = count;
MGLOG_E("MGPipe: verify summary - %llu divergence(s) survived MOBILEGL_PIPE_VERIFY_FATAL=0",
static_cast<unsigned long long>(count));
}
};
VerifyState g_verify;
@@ -645,6 +658,57 @@ namespace MobileGL::MG_Pipe {
MGPipeInputField differing = MGPipeInputField::kFieldCount;
if (!MGPipeVerifyInputs(inputs, g_snapshot, mask, &differing)) ReportDivergence(differing, "entry");
}
#if MOBILEGL_BUILD_DISAGGREGATED
// ---- P7 wave 3 (V1): the compare-at-read oracle inside the server's read_pixels ----
//
// WHAT THE SPLIT ARM FOUND THE FIRST TIME VERIFY AND SPLIT SHARED A BUILD. Every
// glReadPixels on the inproc arm - 8530 reads over 220 of the lane's 1068 entries, and the
// VerifySplitArming. entries on both backends - aborted with
//
// MGPipe: verify read of GetPixelStoreParameters (index 0, 0) differs from the live context
// MGPipe: Fatal{PipeVerifyDiffer, "GetPixelStoreParameters@ReadPixels", verb=6, where=read}
//
// on the server's apply thread, and it was the ONLY field that diverged anywhere in the
// lane. The mechanism is ID-49 and is deliberate: the pack state never shapes the wire
// answer, so MG_Remote/Server/PipeApplier.cpp's read_pixels body writes a NEUTRAL pack
// into the applier's copy of the field (MGPipeApplySetPixelPackState), makes the backend
// read, and restores the pushed one; the client then scatters the tight reply through the
// application's own pack state. The frontend context - this comparator's oracle - still
// holds the application's pack (Alignment 4 by default), so the backend's read of the
// pack half answers {Alignment 1} where the live context says {Alignment 4}. On the
// monolith arm the same neutral window is opened on the frontend context itself
// (GL_Texture.cpp's ScopedNeutralPackState), which is why the monolith lane never saw it.
//
// SO THE ORACLE, NOT THE RULE, IS WHAT CHANGES, AND ONLY IN THAT WINDOW: under a server
// stamp, on the ReadPixels verb, the pack half's expected value is the neutral pack the
// applier reads with, and the UNPACK half and every other field keep the live context as
// their oracle. A server read of the pack half that is neither the application's value
// nor exactly the neutral one is still a divergence and still Fatal.
//
// WHAT THIS GIVES UP IS ONE VALUE, NOT THE FIELD (corrected in P7 wave 3's V1 fix round;
// the first statement of it here said the comparator could no longer see a client that
// pushed a WRONG pack state at all, which is too broad). PipeApplier.cpp's read_pixels
// reads `savedPack` THROUGH THE ACCESSOR - and therefore through this hook - BEFORE it
// installs the neutral value, so the pushed pack has already been compared against the
// live one by then: a correct push is equal and returns, and a wrong push that is not
// exactly the neutral pack fails both compares and is still Fatal. The blind spot is the
// single value {SwapBytes 0, LSBFirst 0, RowLength 0, ImageHeight 0, Skip* 0, Alignment 1},
// and only while the application's own pack differs from it. The client-side scatter that
// consumes the real pack is covered by the readback matrix cases, which compare bytes,
// not fields.
Bool ServerReadsInsideTheNeutralPackWindow(const PipeInputs& self, MGPipeInputField field) {
return field == MGPipeInputField::GetPixelStoreParameters && self.ServerStampedVerb() &&
self.CurrentVerb() == MGPipeVerb::ReadPixels;
}
// The constant itself is MG_Pipe's (MGPipeTypes.h's MGPipeNeutralReadPixelsPack), which is
// the same function MG_Remote/Server/PipeApplier.cpp's read_pixels installs. It used to be
// re-typed here, field for field, to avoid reaching into MG_Remote/Server from MG_Impl -
// but the owner of MGPPixelPackState is MG_Pipe, both sites already include it, and a
// drift between the two spellings would have cost a false divergence rather than a build
// break.
#endif // MOBILEGL_BUILD_DISAGGREGATED
#endif // MOBILEGL_PIPE_VERIFY
} // namespace
@@ -679,13 +743,77 @@ namespace MobileGL::MG_Pipe {
// never recurses (and never reports the inner read against a half-copied scratch).
g_verify.InHook = true;
MGPipeFillAccess::CopyField(g_readScratch, *ctx, field);
// P7 wave 3 (V1): NEGATIVE CONTROL A REACHES THIS ARM TOO. Until this line the knob only
// ever perturbed EntryCompare's snapshot, so every red it could produce said `where=entry`
// and was produced on the CLIENT thread - which left the compare-at-read hook, the arm
// that runs on the server's apply thread and does the whole of the split lane's per-read
// work, with no falsifier at all: a hook that had stopped comparing would have looked
// exactly like a hook with nothing to report. The perturbation goes on the ORACLE, for
// EntryCompare's reason: the arm under test is the stored value, so corrupting THAT would
// be testing the corruption.
//
// WHICH READ THIS BLOCK TURNS RED (V1 fix round 2). The server reads the pack field TWICE
// inside the ReadPixels verb window, and both reads sit inside
// ServerReadsInsideTheNeutralPackWindow below (the verb stamp is up for the whole apply):
// PipeApplier::read_pixels saves the application's pack through the accessor BEFORE it
// installs the neutral one, and the backend's ReadPixels reads the field AFTER. This block
// is what turns the FIRST of them red: at that read `self` still holds the application's
// pack, so the first compare only reaches the window if this perturbation made it differ,
// and inside the window the application's pack against the neutral oracle differs on its
// own. It does nothing for the second read - `self` IS the neutral pack there - which is
// what the re-application inside the window is for.
if (g_verify.Corrupt && *g_verify.Corrupt == field) {
MGPipeApplyVerifyCorruption(g_readScratch, field);
}
const Bool equal = MGPipeInputsFieldEqual(field, self, g_readScratch);
g_verify.InHook = false;
if (equal) return;
#if MOBILEGL_BUILD_DISAGGREGATED
if (ServerReadsInsideTheNeutralPackWindow(self, field)) {
// The live context is the application's pack state and the server is, by ID-49,
// reading with the neutral one; the oracle for THAT HALF is the neutral pack, and
// every other byte of the field still has to match the live context.
g_verify.InHook = true;
PipeInputs::VisitStorage(field, g_readScratch, g_readScratch, [](auto& live, auto&) {
if constexpr (std::is_same_v<std::remove_reference_t<decltype(live)>, PixelStoreParameters[2]>) {
live[0] = MGPipeNeutralReadPixelsPack().Pack;
}
return true;
});
// AND THE CONTROL IS RE-APPLIED, because the overwrite above replaces the pack half
// WHOLESALE - the corruption included, since CorruptStorage perturbs an array's FIRST
// ELEMENT and the pack half is element 0. This is the block for the POST-INSTALL read
// (the backend's ReadPixels, after PipeApplier::read_pixels installed the neutral pack
// into gPipeInputs): there `self` is the neutral pack, the overwrite just made the
// oracle equal to it, and only a perturbation applied AFTER the overwrite can make
// that compare differ - the block above cannot reach it. Without this line the one
// window in which this arm is most load-bearing (the server's read_pixels, 8530 reads
// over 220 of the lane's entries) would be the one window MOBILEGL_PIPE_VERIFY_CORRUPT=
// GetPixelStoreParameters cannot turn red - and a control that accepted ONE report
// would never notice, because the saved-pack read still reports through the block
// above. The VerifySplitReadCorrupted. entries and the CI step therefore require at
// least TWO `where=read` reports in the server half (V1 fix round 2). Measured: 3 on
// DirectGLES (its backend reads the field twice after the install) / 2 on
// DirectVulkan; without this block both drop to 1, without the block above
// DirectVulkan drops to 1 while DirectGLES's two post-install reports keep it at 2 -
// so this block is falsified on both backends and the one above on DirectVulkan.
if (g_verify.Corrupt && *g_verify.Corrupt == field) {
MGPipeApplyVerifyCorruption(g_readScratch, field);
}
const Bool equalToTheNeutralPack = MGPipeInputsFieldEqual(field, self, g_readScratch);
g_verify.InHook = false;
if (equalToTheNeutralPack) return;
}
#endif
MGLOG_E("MGPipe: verify read of %s (index %u, %u) differs from the live context", kMGPipeInputFieldNames[index],
index0, index1);
ReportDivergence(field, "read");
}
// PipeInputs.h. The FATAL=0 summary, written while the role's log is still open:
// MobileGL::Destroy calls this right before MG_Util::Debug::Close() - see VerifyState for why
// the static destructor cannot be the writer.
void MGPipeVerifyFlushSummary() { g_verify.FlushSummary(); }
#endif // MOBILEGL_PIPE_VERIFY
// ---- push on mutation (P1 lane finding F2) ----
@@ -984,8 +1112,24 @@ namespace MobileGL::MG_Pipe {
// below and this aborts by name on the first glBufferData that carries data.
const Uint64 before = MG_Remote::Client::ClientWireRecordsEmitted();
MGPipeEmitResourceSubData(buffer, 0, static_cast<SizeT>(buffer.GetSize()));
if (MG_Remote::Client::ClientWireRecordsEmitted() == before) {
MGLOG_F("MGPipe: Fatal{InitialBytesNotCarried, \"resource_respecify\"} - the "
// EXCEPT ON A SESSION THAT HAS STOPPED CARRYING RECORDS (P12). Once the device-lost
// latch is set every verb is DECLINED by design (CONTRACT-P6 5.3, ClientSession.cpp
// :1837) - the resource_subdata above included - so "it emitted nothing" is the latch
// working, not bytes this client forgot to carry: there is no server left for them to
// exist on, and P12's own exit gate (a) is that killing the server gives a CLEAN
// device-lost latch rather than a corpse. Aborting here turned every clean loss into a
// client SIGABRT at its next glBufferData that carries data.
//
// THE SELF-CHECK KEEPS ITS FULL FORCE ON A LIVE SESSION, which is the case it was
// written for: there, a follow-up that emits nothing is still the R-13.3 gap it names
// - and DeviceLost() is false.
if (MG_Remote::Client::ClientWireRecordsEmitted() == before &&
!MG_Remote::Client::ClientSession::DeviceLost()) {
// UncarriedInitialBytes, not a second word for the same family: WireTables.cpp:418
// already dies of exactly this - a respecify that crossed with no initial bytes -
// under that name, and two words for one family is the vocabulary drift a6
// censused and 5.2's .def exists to bound (P7 wave 0).
MGLOG_F("MGPipe: Fatal{UncarriedInitialBytes, \"resource_respecify\"} - the "
"respecify crossed with initialBytes = nullptr (R-13.3) and the "
"resource_subdata records that were supposed to follow it emitted "
"NOTHING, so %llu bytes of initial content exist on no side of the wire",
@@ -1073,6 +1217,14 @@ namespace MobileGL::MG_Pipe {
#endif
// The application's STAGING store, valid for the duration of the call only.
MGPipeRouteBufferSubDataResident(record, base + (at - offset), length);
// P7 wave 2 package C, OQ-10: `rsd=`, counted HERE rather than at the call
// above so that one glBufferSubData cut into N chunk records counts N. See
// PipeStats.h's CallClass::ResidentSubDataEmissions for why the count is the
// only thing that can tell this arm from the in-place memcpy beside it.
if (MG_Util::PipeStats::Enabled()) {
MG_Util::PipeStats::AddCalls(
MG_Util::PipeStats::CallClass::ResidentSubDataEmissions, 1);
}
}, MGPipeContentChunkCap());
if (!encodable) {
MGLOG_E_ONCE("MGPipe: buffer_subdata_resident range [%llu, +%llu) on buffer %u cannot be encoded",
@@ -1251,6 +1403,14 @@ namespace MobileGL::MG_Pipe {
// P5e (ra), CONTRACT-P5E §2.5. Declared in PipeMutation.h; see there for why it is not a
// ClientSession call at the GL entry point.
void MGPipeClientFlush() {
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
if (MG_Remote::Client::RunsAsTheServerRole()) return;
if (MG_Remote::Client::ClientSession* session = MG_Remote::Client::ClientSession::Active()) {
session->Flush();
}
}
void MGPipeClientFinish() {
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
if (MG_Remote::Client::RunsAsTheServerRole()) return;
@@ -1454,79 +1614,44 @@ namespace MobileGL::MG_Pipe {
// above: with a dependency unmet the client emits NOTHING for that family and the legacy
// pull path runs untouched, on both sides of the boundary.
//
// THE TABLE IS WRITTEN ONCE, HERE, and every one of its rows is the client mirror of the
// refusal Espryt already implements, bit for bit and non-transitively - the two must say
// the SAME thing, because a client that withheld more than the server refuses would
// leave the server's handle arm live with no records to read, and a client that withheld
// less is the defect above.
struct P4aFamilyDependencyRow {
Uint64 Family; // exactly one bit, and it is one of kMGPipeP4aFamilySubsystems
Uint64 Requires; // the bits MOBILEGL_PIPE_PUSH must ALSO carry for it to be live
};
inline constexpr P4aFamilyDependencyRow kMGPipeP4aFamilyDependencies[] = {
// BIT 9 REQUIRES BIT 10. Every MGPSurface::Res in a set_framebuffer_state record
// names a Texture or a Renderbuffer handle, and only bit 10 populates those two slot
// tables (Managers.cpp ResolveFramebufferSubsystemArm).
{kMGPipeSubsystemFramebuffer, kMGPipeSubsystemTextureResources},
// BIT 10 REQUIRES BIT 7 - a buffer texture's MGPResourceDesc::BufferForTexBuffer
// names a Buffer handle and only bit 7 puts twins in the resource slot table (D-D1,
// ResolveTextureResourceSubsystemArm's first row) - AND BIT 11, which is D-K2's
// FOURTH row (ID-14/ID-15): MGPTextureParams::BuiltinSampler is a SamplerCso HANDLE,
// only bit 11 mints sampler CSOs (c0b's four unconditional mints deliberately
// exclude it), and the applier's verdict for a null one is Fatal{ProtocolCorruption}
// rather than a decline. The brief's original "bit 10 without 11 is fine" is
// WITHDRAWN for P4a as built.
{kMGPipeSubsystemTextureResources,
kMGPipeSubsystemResources | kMGPipeSubsystemSamplers},
// BIT 11 REQUIRES BIT 10. Every MGPBoundView::Texture and every MGPImageView::Res
// names a Texture handle and only bit 10 populates that slot table; without it every
// per-unit lookup would miss and the walk would `continue` WITHOUT unbinding
// (ResolveSamplerSubsystemArm). With the row above this is SYMMETRIC: bits 10 and 11
// are one arm with two switches, and the only two masks that reach either handle arm
// are "both set" and "neither set".
{kMGPipeSubsystemSamplers, kMGPipeSubsystemTextureResources},
// BIT 12 DEPENDS ON NOTHING, and that is a ROW rather than an absence so the table
// covers the four families exhaustively (the static_assert below): a ShaderCso handle
// names no texture and no buffer, the archive rides beside the record as a companion
// pointer, and the extra inputs the server specialises on are read from state the
// backend already holds (ResolveProgramSubsystemArm).
{kMGPipeSubsystemPrograms, 0},
};
// THE MIRROR PAIRS THAT STAY FINE, said out loud rather than left as an absence, because
// an unreachable branch that says something different is how the reachable one drifts
// (Managers.cpp's own words at :2377-2381) - and because the table is only trustworthy if
// what it does NOT contain was decided rather than forgotten:
// - bit 10 set, bit 9 clear: FINE. The legacy FBO sync reaches the texture twin through
// SyncTextureObjectToBackend, which dispatches to the handle arm by itself.
// - bit 11 set, bit 9 clear: FINE, for the same reason - a sampler view names a texture,
// never a framebuffer.
// - bit 7 set, bit 10 clear: FINE, and it is P3a's shipped configuration.
// - bit 12 set with any or none of 9/10/11: FINE, per the last row.
// - bit 10 set, bit 11 clear (and its mirror) is NOT fine and is the row above; this is
// the one sentence in the brief that P4a as built withdrew.
// THE TABLE IS WRITTEN ONCE, IN MG_Pipe/SubsystemDeps.def, and every one of its rows is
// the client mirror of the refusal Espryt already implements, bit for bit and
// non-transitively - the two must say the SAME thing, because a client that withheld more
// than the server refuses would leave the server's handle arm live with no records to
// read, and a client that withheld less is the defect above.
//
// THIS FILE USED TO HOLD ITS OWN COPY of the four P4a rows (P3b/P4b R-5 found six
// statements of one rule and two of them drifted). The rows moved to the .def with R-5
// and this reader was left pointing at the copy deliberately, because PipeFill.cpp is the
// contract package's file for the phase; wave 2-D package D3 switches it over.
//
// THE ARGUMENT IS MASKED TO kMGPipeP4aFamilySubsystems, AND THAT IS THE WHOLE DIFFERENCE
// BETWEEN A REWRITE AND A BEHAVIOUR CHANGE. The .def carries SIX rows; the table deleted
// from here carried FOUR, and this predicate is `wants()`'s fourth conjunct for P4a's
// families only. Reading all six unmasked would make the CLIENT withhold bit 8's
// vertex-input emissions at a mask with bit 7 clear - P3a's own rule, which this gate has
// never narrowed and which TextureEmitTest's
// EveryDKTwoDependencyRowGatesItsOwnFamilyAndTheMirrorPairsStayLive pins by name ("bit 8
// alone, with bit 7 clear: P3a's own rule, which this table must not touch"). Bit 13 is
// not narrowed here either; it asks the table for itself in P5eFamilyIsLive below,
// because bit 13's conjunct is bit 13's own. So the ROWS now come from one place and WHO
// each gate speaks for is still each gate's.
//
// The P4a-specific COVERAGE assertion stays for the same reason: a fifth P4a family
// without a row in the .def has to be a compile error, not a silent "depends on nothing".
constexpr Uint64 P4aFamilyDependencyBits(Uint64 subsystem) {
Uint64 required = 0;
for (const P4aFamilyDependencyRow& row : kMGPipeP4aFamilyDependencies) {
if ((subsystem & row.Family) != 0) required |= row.Requires;
}
return required;
return MGPipeSubsystemRequires(subsystem & kMGPipeP4aFamilySubsystems);
}
// The table covers the four families this phase migrates and nothing else, so a fifth
// family added to kMGPipeP4aFamilySubsystems without a row here does not silently inherit
// "depends on nothing".
constexpr Uint64 P4aFamilyDependencyTableCoverage() {
Uint64 covered = 0;
for (const P4aFamilyDependencyRow& row : kMGPipeP4aFamilyDependencies) covered |= row.Family;
for (const MGPipeSubsystemDependencyRow& row : kMGPipeSubsystemDependencies) covered |= row.Family;
return covered;
}
static_assert(P4aFamilyDependencyTableCoverage() == kMGPipeP4aFamilySubsystems,
"every P4a family needs a D-K2 dependency row, even an empty one");
static_assert((P4aFamilyDependencyTableCoverage() & kMGPipeP4aFamilySubsystems) ==
kMGPipeP4aFamilySubsystems,
"every P4a family needs a D-K2 dependency row in MG_Pipe/SubsystemDeps.def, "
"even an empty one");
static_assert(P4aFamilyDependencyBits(kMGPipeSubsystemFramebuffer) ==
kMGPipeSubsystemTextureResources,
"bit 9 requires bit 10");
@@ -1582,14 +1707,18 @@ namespace MobileGL::MG_Pipe {
// server whose four binding-point walks still read the frontend, so a record sent to it
// would be stored and never looked at while the client latched its suppressor.
//
// AND BIT 13 REQUIRES BIT 7, for bit 11's reason one family over: every
// MGPBufferRange::Res names a Buffer handle and only bit 7 puts one in the resource
// slot table, so without it every EnsureBufferResourceForHandle on the server would
// mint a twin with no record behind it. Withholding the whole family is the safe
// direction - the legacy frontend walk runs untouched on both sides.
// AND BIT 13'S OWN DEPENDENCY ROW IS READ FROM MG_Pipe/SubsystemDeps.def, exactly as
// P4a's four are (P3b/P4b R-5, switched over by wave 2-D package D3). It used to be the
// hand-coded `(pushMask & kMGPipeSubsystemResources) == 0` below - the sixth statement of
// the rule, and the one that sat fifteen lines from the block stating the other three,
// which is how "THREE OF THEM" survived two phases. Its reason is the table's: every
// MGPBufferRange::Res names a Buffer handle and only bit 7 puts one in the resource slot
// table, so without it every EnsureBufferResourceForHandle on the server would mint a
// twin with no record behind it. Withholding the whole family is the safe direction - the
// legacy frontend walk runs untouched on both sides.
Bool P5eFamilyIsLive(Uint64 subsystem, Uint64 pushMask) {
if ((subsystem & kMGPipeSubsystemBufferBindings) == 0) return true;
if ((pushMask & kMGPipeSubsystemResources) == 0) return false;
if (!MGPipeSubsystemDependenciesAreSet(kMGPipeSubsystemBufferBindings, pushMask)) return false;
#if MOBILEGL_BUILD_DISAGGREGATED
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
return MG_Remote::Client::CapsMirrorInstance().ServerConsumes(
@@ -1699,6 +1828,11 @@ namespace MobileGL::MG_Pipe {
*entry = Entry{};
}
void Clear() {
for (auto& table : m_kinds) table.clear();
m_band.clear();
}
private:
struct Entry {
Uint32 Gen = 0;
@@ -1763,6 +1897,15 @@ namespace MobileGL::MG_Pipe {
PublicationLatch().NoteUnpublished(kind, handle);
}
void MGPipeForgetAllPublications() {
PublicationLatch().Clear();
// The two families whose death asks their own record rather than the latch: vertex
// elements (MGPipeEmitVertexElementsDestroyAndFree's RecordIsPublished) and buffers
// (MGPipeEmitResourceDestroyAndFree's WasPublished).
MGPipeVertexInputEmitterInstance().ForgetAllRecords();
MGPipeResourceTrackerInstance().ForgetAllPublished();
}
// THE GATE ITSELF, AS AN OBSERVABLE (ID-39, widened by S-3 / ID-41). Every P4a birth hook
// below and every `wants()` row in the walk resolve through FamilyIsLive /
// P4aFamilyHasItsConsumer / P4aFamilyDependenciesAreSet, and none of the three is reachable
@@ -1871,8 +2014,11 @@ namespace MobileGL::MG_Pipe {
//
// BACKEND-NEUTRAL FROM THE FIRST COMMIT, which is the whole point: before P3a's C-1 fix
// the only thing that ever returned a VertexElementsCso slot was DirectGLES'
// StateObjectDeathOps table, so under a backend that installs none every VAO leaked a slot
// and a ~1.3 KB applier record for the life of the process. P4a mints SIX kinds and there
// StateObjectDeathOps table, so under a backend that installed none every VAO leaked a slot
// and a ~1.3 KB applier record for the life of the process. It stays the point now that
// BOTH backends install one (P7 wave 2 package C gives Magma its own, CONTRACT-P7 §5.5):
// Magma's table EMITS the death record and frees nothing, so the slot still comes back
// from here and from nowhere else. P4a mints SIX kinds and there
// is no intermediate state in which a backend table is the only path for any of them.
//
// THE THREE-STEP ORDER IS FIXED and each position is load-bearing (see PipeMutation.h):
+9 -3
View File
@@ -61,9 +61,15 @@ namespace MobileGL::MG_Pipe {
// 47 DirectGLES cases at ID-41). With them the four families emit NOTHING in that state and
// the legacy pull path runs exactly as it does on a pull build.
//
// D-K2's TABLE IS IN PipeFill.cpp, ONCE: bit 9 requires bit 10, bit 10 requires bits 7 and
// 11, bit 11 requires bit 10, bit 12 depends on nothing - the client mirror, bit for bit, of
// the four `Resolve<Family>SubsystemArm()` refusals in DirectGLES/Managers.cpp.
// D-K2's ROWS ARE IN MG_Pipe/SubsystemDeps.def, ONCE (P3b/P4b R-5), and this comment no
// longer restates them - restating them here is half of how the rule ended up with six
// statements, two of which were wrong. PipeFill.cpp's kMGPipeP4aFamilyDependencies is still
// the client's own copy and is still the bit-for-bit mirror of the four
// `Resolve<Family>SubsystemArm()` refusals in DirectGLES/Managers.cpp; switching it to read
// the .def is the integrator's one-line change (PipeFill.cpp is the contract package's file
// for the whole phase, so the emitter packages do not edit it).
// MG_Test/Backend/DirectGLES/SubsystemDepsTest.cpp drives BOTH readers at every interesting
// mask and compares them against the .def, so the two cannot drift in the meantime.
//
// It is exported for the unit gate and for no other caller: the gate itself is
// FamilyIsLive() inside PipeFill.cpp, every birth hook and every `wants()` row resolves

Some files were not shown because too many files have changed in this diff Show More